# Model Type Support In SwarmUI | Model | Architecture | Year | Author | Scale | Censored? | Quality/Status | | ---- | ---- | ---- | ---- | ---- | ---- | ---- | [Stable Diffusion XL](#stable-diffusion-xl) | unet | 2023 | Stability AI | 2B | Partial | Old but some finetunes remain popular | [SD1 and SDXL Turbo Variants](#sd1-and-sdxl-turbo-variants) | unet | 2023 | Stability AI and others | 2B | Partial | Outdated | [Stable Diffusion 3](#stable-diffusion-3) | MMDiT | 2024 | Stability AI | 2B | Yes | Outdated, prefer 3.5 | [Stable Diffusion 3.5 Large](#stable-diffusion-35-large) | MMDiT | 2024 | Stability AI | 8B | Partial | Outdated, Good Quality for its time | [Stable Diffusion 3.5 Medium](#stable-diffusion-35-medium) | MMDiT | 2024 | Stability AI | 2B | Partial | Outdated, Good Quality for its time | [Flux.1](#black-forest-labs-flux1-models) | MMDiT | 2024 | Black Forest Labs | 12B | Partial | Outdated, High Quality for its time | [Flux.2](#flux-2) | MMDiT | 2025 | Black Forest Labs | 4B, 9B, 32B | Minimal | Recent, Incredible Quality, choice of speed or quality preference | [Chroma](#chroma) | MMDiT | 2025 | Lodestone Rock | 8.9B | No | Recent, Decent Quality | [Chroma Radiance](#chroma-radiance) | Pixel MMDiT | 2025 | Lodestone Rock | 8.9B | No | Recent, Bad Quality (WIP) | [Lumina 2.0](#lumina-2) | NextDiT | 2025 | Alpha-VLLM | 2.6B | Partial | Recent, Passable Quality | [Qwen Image](#qwen-image) | MMDiT | 2025 | Alibaba-Qwen | 20B | Minimal | Recent, Great Quality, very memory intense | [Hunyuan Image 2.1](#hunyuan-image-21) | MMDiT | 2025 | Tencent | 17B | No | Recent, Great Quality, very memory intense | [Z-Image](#z-image) | S3-DiT | 2025 | Tongyi MAI (Alibaba) | 6B | No | Recent, Great Quality, lightweight | [Kandinsky 5](#kandinsky-5) | DiT | 2025 | Kandinsky Lab | 6B | No | Recent, Decent Quality | [Anima](#anima) | DiT | 2026 | Circlestone Labs | 2B | WTF | Modern, very small, decent for anime | [HiDream O1](#hidream-o1) | "Pixel UiT" | 2026 | HiDream | 8B | Minimal | Modern, intelligent, fast, decent quality | [Ideogram 4](#ideogram-4) | DiT | 2026 | Ideogram AI | 9B | Yes | Modern, advanced on input understanding | [Krea 2](#krea-2) | DiT | 2026 | Krea AI | 12B | Yes | Modern, extremely smart and great quality | [Boogu](#boogu) | MMDiT | 2026 | Boogu | 10B | Minimal | Modern, fast | [Qwen Image 2.1](#qwen-image-21) | MMDiT | 2026 | Alibaba-Qwen | 7B | No | Modern, creative, slow | [Ming Image](#ming-image) | DiT | 2026 | InclusionAI | 6B | Minimal | Modern, intelligent but ugly | [Flux 3 Image](#flux-3-image) | DiT | 2026 | Black Forest Labs | 5.6B | Yes | Modern, intelligent but poor details | [MOSS Image](#moss-image) | DiT | 2026 | OpenMOSS | 5B | No | Modern, extremely capable | [Fibo 2](#bria-fibo-2) | DiT | 2026 | Bria | 8B | Minimal | Modern, meh quality | [MiniMax H3 Image](#minimax-h3-image) | DiT | 2026 | MiniMax AI | 20B | Minimal | Modern, intelligent but poor details | Old or bad options also tracked listed via [Obscure Model Support](/docs/Obscure%20Model%20Support.md): | Model | Architecture | Year | Author | Scale | Censored? | Quality/Status | | ---- | ---- | ---- | ---- | ---- | ---- | ---- | [Stable Diffusion v1 and v2](/docs/Obscure%20Model%20Support.md#stable-diffusion-v1-and-v2) | unet | 2022 | Stability AI | 1B | No | Outdated | [Stable Diffusion v1 Inpainting Models](/docs/Obscure%20Model%20Support.md#stable-diffusion-v1-inpainting-models) | unet | 2022 | RunwayML | 1B | No | Outdated | [Segmind SSD 1B](/docs/Obscure%20Model%20Support.md#segmind-ssd-1b) | unet | 2023 | Segmind | 1B | Partial | Outdated | [Stable Cascade](/docs/Obscure%20Model%20Support.md#stable-cascade) | unet cascade | 2024 | Stability AI | 5B | Partial | Outdated | [PixArt Sigma](/docs/Obscure%20Model%20Support.md#pixart-sigma) | DiT | 2024 | PixArt | 1B | ? | Outdated | [AuraFlow](/docs/Obscure%20Model%20Support.md#auraflow) | MMDiT | 2024 | Fal.AI | 6B | Yes | Outdated | [Nvidia Sana](/docs/Obscure%20Model%20Support.md#nvidia-sana) | DiT | 2024 | NVIDIA | 1.6B | No | Just Bad | [Nvidia Cosmos Predict2](/docs/Obscure%20Model%20Support.md#cosmos-predict2) | DiT | 2025 | NVIDIA | 2B/14B | Partial | Just Bad | [HiDream i1](/docs/Obscure%20Model%20Support.md#hidream-i1) | MMDiT | 2025 | HiDream AI (Vivago) | 17B | Minimal | Good Quality, lost community attention | [OmniGen 2](/docs/Obscure%20Model%20Support.md#omnigen-2) | MLLM | 2025 | VectorSpaceLab | 7B | No | Modern, Decent Quality, quickly outclassed | [Ovis](/docs/Obscure%20Model%20Support.md#ovis) | MMDiT | 2025 | AIDC-AI (Alibaba) | 7B | No | Passable quality, but outclassed on launch | [LongCat-Image](/docs/Obscure%20Model%20Support.md#longcat-image) | MMDiT | 2025 | LongCat | 6B | No | Passable quality, but outclassed on launch | [Zeta Chroma](/docs/Obscure%20Model%20Support.md#zeta-chroma) | Pixel S3-DiT | 2026 | Lodestone Rock | 6B | No | Modern, Pixel-space Z-Image variant | [PixelDiT](/docs/Obscure%20Model%20Support.md#pixeldit) | Pixel DiT | 2026 | NVIDIA | 1.3B | Minimal | Modern, fast, pixel-space, but very bad relative quality on launch | [MageFlow](/docs/Obscure%20Model%20Support.md#mageflow) | MMDiT | 2026 | Microsoft | 4B | No | Very bad quality compared to contemporary models | [ERNIE](/docs/Obscure%20Model%20Support.md#ernie) | DiT | 2026 | Baidu | 8B | Minimal | Modern, intelligent, good quality, fast | [Lens](/docs/Obscure%20Model%20Support.md#lens) | MMDiT | 2026 | Microsoft | 4B | Minimal | Modern, lightweight, eh quality | [SenseNova-U1](/docs/Obscure%20Model%20Support.md#sensenova) | "Pixel UiT" | 2026 | SenseNova | 8B | Minimal | Modern, very intelligent but bad details | - **Architecture** is the fundamental machine learning structure used for the model, UNet's were used in the past but DiT (Diffusion Transformers) are the modern choice - **Scale** is how big the model is - "B" for "Billion", so for example "2B" means "Two billion parameters". - One parameter is one number value, so for example in fp16 (16 bit, ie 2 bytes per number), a 2B model is 4 gigabytes. In fp8 (8 bit, ie 1 byte per number), a 2B model is 2 gigabytes. - If you often use fp8 or q8 models, just read the "B" as "gigabytes" for a good approximation - **Censored?** is tested by generating eg "a photo of a naked woman" on the model. - This test only refers to the base models, finetunes can add nudity and other "risque" content back in. - Most base models will not generate genitalia, and have limited quality with other body parts and poses. Every popular model has finetunes available to add those capabilities, if you want them. - Sometimes it's not even intentional censorship, just the simple fact that broad base models aren't good at any one thing - so, again, content-specific finetunes fix that. - Model censorship can take other forms (eg does it recognize names of celebrities/artists/brands, can it do gore, etc.) so if a model sounds right to you you may want do your own testing to see if it's capable of the type of content you like - "No" means it generates what was asked, - "Minimal" means it's eg missing genitals but otherwise complete, - "Partial" means it's clearly undertrained at NSFW content (eg difficult to prompt for or poor quality body) but doesn't explicitly refuse, - "Yes" means it's entirely incapable or provides an explicit refusal response, - "WTF" means it's the opposite of censored, may generate inappropriate content even without being asked. - **Quality/Status** is a loose vibe-based metric to imply whether it's worth using in the current year or not. - Video models are in [Video Model Support](/docs/Video%20Model%20Support.md) - Audio models are in [Audio Model Support](/docs/Audio%20Model%20Support.md) # Current Recommendations Image model(s) most worth using, as of September 2026: - Krea 2 is the champion of image generation. - Flux.2 Klein is pretty great too, particularly for editing. - Qwen Image 2.1 is probably better for edit than Klein. - Ideogram or Ming Image may be worth experimenting with if you like getting advanced, as they allow JSON inputs with structured bounding boxes. # General Info - Swarm natively supports `.safetensors` format models with [ModelSpec](https://github.com/Stability-AI/ModelSpec) metadata - can also import metadata from some legacy formats used by other UIs (auto webui thumbnails, matrix jsons, etc) - can also fallback to a `.swarm.json` sidecar file for other supported file formats - Swarm can load other model file formats, see [Alternative Model Formats](#alternative-model-formats) - Notably, *quantization* technique formats. "Quantization" means shrinking a model to use lower memory than is normally reasonable. - Normal sizes are named like "BF16", "FP16", "FP8", ... ("BF"/"FP" prefixes are standard formats) - There are also special-but-still-normal types, such as "mxfp8", "nvfp4" (these are not proper standard types, but are nonetheless core supported) - Quantized sizes have names like "NF4", "Q4_K_M", "Q8", "SVDQ-4", "Int-4", ("Q" means quantized, but there are technique-specific labels) - "int8" format models are natively supported, with some temporary limitations in current versions - [BnB NF4](#bits-and-bytes-nf4-format-models) (not recommended, quantization technique) - [GGUF](#gguf-quantized-models) (recommended, good quality quantization technique, slower speed) - [Nunchaku](#nunchaku-mit-han-lab) (very recommended, great quality high speed quantization technique) - [TensorRT](#tensorrt) (not recommended, speedup technique) # Image Models - The early image demos included below are mini-grids of seeds `1, 2, 3` res 1:1 of the prompt `wide shot, photo of a cat with mixed black and white fur, sitting in the middle of an open roadway, holding a cardboard sign that says "Meow I'm a Cat". In the distance behind is a green road sign that says "Model Testing Street".` ran on each model. - This prompt is designed to require (1) multiple complex components (2) photorealism (3) text (4) impossible actions (cat holding a sign - Most models get very confused how to do this). - In 2026, models became sufficiently capable as to irrelevantize this test. Therefore, a tougher prompt is used for later prompts: a single 21:9 run of seed `1` with this prompt: ``` Four panels in horizontal parallel. Panel one (leftmost): wide shot, photo of a cat with mixed black and white fur, sitting in the middle of an open roadway, holding a cardboard sign that says "Meow I'm a Cat". In the distance behind is a green road sign that says "Model Testing Street". Panel two: anime art, Frieren art style, 1girl, a 21 year old magical princess. Her face resembles Hatsune Miku. She wears a purple hat, a large flowing white dress, with black and purple accents detailing her arm bracers, leggings, and other bits n bobs. Green forest background. Panel three: Spongebob Squarepants art style, showing Patrick Star shaking hands with Homer Simpson (still in the Spongebob style) Panel four (rightmost): real photo, close up, Ben Affleck's face. He is a grizzled soldier, with a tough expression, staring dead at the camera. Across the whole bottom of the image: Simple white text with black dropshadow, reads clearly: "This prompt is intended to test the model's range, capabilities, and intelligence." The second line reads "This text should be readable, and the characters and styles in each panel should match the prompt." ``` - For all models, "standard parameters" are used. - Steps is set to 20 except for Turbo models. Turbo models are ran at their standard fast steps (usually 4 or 8). - CFG is set appropriate to the model. - Any separated prompt upsampler LLMs are *not* run. - All generations are done on the base model of the relevant class, not on any finetune/lora/etc. Finetunes are likely to significantly change the qualitative capabilities, but unlikely to significantly change general ability to understand and follow prompts. - This is not a magic perfect test prompt, just a decent coverage of range to showcase approximately what you can expect from the model in terms of understanding and handling challenges. - You could make a point that maybe I should have set CFG different or used a sigma value or changed up prompt phrasing or etc. and get better quality - this test intentionally uses very bland parameters to maximize identical comparison. Keep in mind that you can get better results out of a model by fiddling parameters. - You'll note models started being able to do decently well on this test in late 2024. Older models noticeable fail at the basic requirements of this test. # Stable Diffusion XL ![img](/docs/images/models/sdxl.jpg) SDXL models work as normal, with the bonus that by default enhanced inference settings will be used (eg scaled up rescond). Additional, SDXL-Refiner architecture models can be inferenced, both as refiner or even as a base (you must manually set res to 512x512 and it will generate weird results). ### SDXL Controlnets - There are official SDXL ControlNet LoRAs from Stability AI [here](https://huggingface.co/stabilityai/control-lora) - and there's a general collection of community ControlNet models [here](https://huggingface.co/lllyasviel/sd_control_collection/tree/main) that you can use. # SD1 and SDXL Turbo Variants Turbo, LCM (Latent Consistency Models), Lightning, etc. models work the same as regular models, just set `CFG Scale` to `1` and: - For Turbo, `Steps` to `1` Under the `Sampling` group set `Scheduler` to `Turbo`. - For LCM, `Steps` to `4`. Under the `Sampling` group set `Sampler` to `lcm`. - For lightning, (?) # Stable Diffusion 3 ![img](/docs/images/models/sd3m.jpg) Stable Diffusion 3 Medium is supported and works as normal. By default the first time you run an SD3 model, Swarm will download the text encoders for you. Under the `Sampling` parameters group, a parameter named `SD3 TextEncs` is available to select whether to use CLIP, T5, or both. By default, CLIP is used (no T5) as results are near-identical but CLIP-only has much better performance, especially on systems with limited resources. Under `Advanced Sampling`, the parameter `Sigma Shift` is available. This defaults to `3` on SD3, but you can lower it to around ~1.5 if you wish to experiment with different values. Messing with this value too much is not recommended. For upscaling with SD3, the `Refiner Do Tiling` parameter is highly recommended (SD3 does not respond well to regular upscaling without tiling). ### Stable Diffusion 3.5 Large ![img](/docs/images/models/sd35l.jpg) - Stable Diffusion 3.5 Large is supported and works as normal, including both normal and Turbo variants. - The [TensorArt 3.5L TurboX](https://huggingface.co/tensorart/stable-diffusion-3.5-large-TurboX/tree/main) works too, just set `CFG Scale` to `1`, and `Steps` to `8`, and advanced `Sigma Shift` to `5` - They behave approximately the same as the SD3 Medium models, including same settings and all, other than harsher resource requirements and better quality. - You can also use [GGUF Versions](#gguf-quantized-models) of the models. - SD3.5 Large or LargeTurbo ### Stable Diffusion 3.5 Medium ![img](/docs/images/models/sd35m.jpg) - Stable Diffusion 3.5 Medium is supported and works as normal. - They behave approximately the same as the SD3 Medium models, including same settings and all. - You can also use [GGUF Versions](#gguf-quantized-models) of the models. - SD3.5 Medium - SD 3.5 Medium support resolutions from 512x512 to 1440x1440, and the model metadata of the official model recommends 1440x1440. However, the official model is not good at this resolution. You will want to click the `☰` hamburger menu on a model, then `Edit Metadata`, then change the resolution to `1024x1024` for better results. You can of course set the `Aspect Ratio` parameter to `Custom` and the edit resolutions on the fly per-image. # Black Forest Labs' Flux.1 Models ### Flux Dev ![img](/docs/images/models/flux-dev.jpg) ### Flux Schnell ![img](/docs/images/models/flux-schnell.jpg) ### Install - Black Forest Labs' Flux.1 model is fully supported in Swarm - **Recommended:** for best performance on modern nvidia cards, use Nunchaku models. - These run twice as fast as the next best speed option (fp8) while using less memory too (close to gguf q4) - Flux dev - Flux Schnell - Use "fp4" for Blackwell (eg RTX 5090) or newer cards, use "int4" for anything older (4090, 3090, etc.) - See the [Nunchaku Support](#nunchaku-mit-han-lab) section for more info on this format - **Recommended:** use the [GGUF Format Files](#gguf-quantized-models) (best for most graphics cards) - Flux Schnell - Flux Dev - `Q6_K` is best accuracy on high VRAM, but `Q4_K_S` cuts VRAM requirements while still being very close to original quality, other variants shouldn't be used normally - Goes in `(Swarm)/Models/diffusion_models` - After adding the model, refresh the list, then you may need to click the `☰` hamburger menu on the model, then `Edit Metadata` and set the `Architecture` to `Flux Dev` or `Flux Schnell` as relevant, unless it detects correctly on its own. - **Alternate:** the simplified fp8 file (best on 3090, 4090, or higher tier cards): - Dev - Schnell - goes in your regular `(Swarm)/Models/Stable-Diffusion` dir - **or, not recommended:** You can download BFL's original files: - Download "Schnell" (Turbo) from - Or "Dev" (non-Turbo) from - Goes in `(Swarm)/Models/diffusion_models` - Required VAE & TextEncoders will be autodownloaded if you do not already have them, you don't need to worry about those. ### Parameters - **CFG Scale:** For both models, use `CFG Scale` = `1` (negative prompt won't work). - For the Dev model, there is also a `Flux Guidance Scale` parameter under `Sampling`, which is a distilled embedding value that the model was trained to use. - Dev can use some slightly-higher CFG values (allowing for negative prompt), possibly higher if you reduce the Flux Guidance value and/or use Dynamic Thresholding. - **Sampler:** Leave it at default (Euler + Simple) - **Steps:** For Schnell use Steps=4 (or lower, it can even do 1 step), for Dev use Steps=20 or higher - **Resolution:** It natively supports any resolution up to 2 mp (1920x1088), and any aspect ratio thereof. By default will use 1MP 1024x1024 in Swarm. You can take it down to 256x256 and still get good results. - You can mess with the resolution quite a lot and still get decent results. It's very flexible even past what it was trained on. - You _can_ do a refiner upscale 2x and it will work but take a long time and might not have excellent quality. - Enable `Refiner Do Tiling` for any upscale target resolution above 1536x1536. ### Performance - Flux is best on a very high end GPU (eg 4090) for now. It is a 12B model. - Smaller GPUs can run it, but will be slow. This requires a lot of system RAM (32GiB+). It's been shown to work as low down as an RTX 2070 or 2060 (very slowly). - On a 4090, schnell takes about 4/5 seconds to generate a 4-step image, very close to SDXL 20 steps in time, but much higher quality. - By default swarm will use fp8_e4m3fn for Flux, if you have a very very big GPU and want to use fp16/bf16, under `Advanced Sampling` set `Preferred DType` to `Default (16 bit)` ### Flux.1 Tools - The Flux.1 Tools announced [here by BFL](https://blackforestlabs.ai/flux-1-tools/) are supported in SwarmUI - For "**Redux**", a Flux form of image prompting: - Download [the Redux model](https://huggingface.co/black-forest-labs/FLUX.1-Redux-dev/blob/main/flux1-redux-dev.safetensors) to `(SwarmUI)/Models/style_models` - (Don't worry about sigclip, it is automanaged) - Drag an image to the prompt area - On the top left, find the `Image Prompting` parameter group - Select the `Use Style Model` parameter to the Redux model - There's an advanced `Style Model Apply Start` param to allow better structural control from your text prompt - set to 0.1 or 0.2 or so to have the text prompt guide structure before redux takes over styling - at 0, text prompt is nearly ignored - The advanced `Style Model Merge Strength` param lets you partial merge the style model against the nonstyled input, similar to Multiply Strength - The advanced `Style Model Multiply Strength` param directly multiplies the style model output, similar to Merge Strength - For "**Canny**" / "**Depth**" models, they work like regular models (or LoRAs), but require an Init Image to function. - Goes in the regular `diffusion_models` or lora folder depending on which you downloaded. - You must input an appropriate image. So eg for the depth model, input a Depth Map. - You can use the controlnet parameter group to generate depth maps or canny images from regular images. - (TODO: Native interface to make that easier instead of janking controlnet) - Make sure to set Creativity to `1`. - This is similar in operation to Edit models. - For "**Fill**" (inpaint model), it works like other inpaint models. - It's a regular model file, it goes in the regular `diffusion_models` folder same as other flux models. - "Edit Image" interface encouraged. - Mask a region and go. - Creativity `1` works well. - Larger masks recommended. Small ones may not replace content. - Boosting the `Flux Guidance Scale` way up to eg `30` may improve quality - For "**Kontext**" (edit model), it works like other edit models. - Model download here - Or the nunchaku version here - Or the official BFL 16 bit upload - Or some GGUFs here - It's a regular model file, it goes in the regular `diffusion_models` folder same as other flux models. - You will have to manually edit the architecture to be `Flux.1 Kontext Dev`, it misdetects by default - Click the `☰` hamburger menu on the model, then `Edit Metadata`, then select the `Architecture` as `Flux.1 Kontext Dev`, then hit save - Paste images into the prompt box to serve as the reference images it will use to generate. - If you have an init image and no reference images, the init image will be used. - Swarm will automatically keep the size of the image correct for Kontext input, but make sure your aspect ratio is matched. - Kontext can take as many images as you want, but the way this works on the inside is a bit hacky and limited quality. - Prompt should describe a *change* to make to the image. - BFL published an official prompting guide here, following it carefully is recommended: - If you want to use the **ACE Plus** Models (Character consistency) - Download the LoRAs from https://huggingface.co/ali-vilab/ACE_Plus/tree/main and save as normal loras - Enable the Flux Fill model, enable the LoRA you chose - Set `Flux Guidance Scale` way up to `50` - Open an image editor for the image you want to use as an input (Drag to center area, click Edit Image) - set `Init Image Creativity` to 1 (max) - Change your `Resolution` parameters to have double the `Width` (eg 1024 input, double to 2048) - Add a Mask, draw a dot anywhere in the empty area (this is just a trick to tell the editor to automask all the empty area to the side, you don't need to mask it manually) - Type your prompt, hit generate # Flux 2 ![img](/docs/images/models/flux2.jpg) - Black Forest Labs' [Flux.2 Models](https://bfl.ai/blog/flux-2) are supported in SwarmUI - The main "Dev" model is an extremely massive model (32B diffusion model, 24B text encoder) that will demand significant RAM availability on your PC. - This can easily fill up 128 gigs of system RAM in usage, but does still work on 64 gig systems. Lower than 64 may not be possible, or may require heavily using swapfile. - The smaller [Klein model](#flux2-klein) is preferred for more normal PC hardware. - Download the standard FP8 model here [silveroxides/FLUX.2-dev-fp8_scaled]() - Or GGUF version here [city96/FLUX.2-dev-GGUF]() - Goes in `diffusion_models` folder - There's also a turbo model here [silveroxides/flux2-dev-turbo-fp8mixed.safetensors]() - or as a lora [fal/Flux_2-Turbo-LoRA_comfyui.safetensors]() - Use scheduler `Align Your Steps` at exactly 8 steps to select the special Flux2-dev-turbo custom scheduler. Difference is minimal though. - The VAE is a brand new 16x16 downsample VAE with 128 channels. It will be autodownloaded. - You can swap it for this one [CabalResearch/Flux2VAE-Anime-Decoder-Tune]() which is a finetune to reduce detail artifacting. - The Text Encoder is 24B Mistral Small 3.2 (2506). It will be autodownloaded. - There's a GGUF about half the size here [mcmonkey/flux2MistralGGUF]() - Select via the advanced `Mistral Model` parameter - **Parameters:** - **Prompt:** Prompting guide from the model creators here - Their docs recommend a JSON structured prompting, but this does not seem all that much more useful in practice, and the JSON syntax in a prompt may trigger parsing issues - They used a powerful LLM for inputs, allow for multiple languages and a variety of ways of phrasing/formatting text to work out - **Resolution:** Flux2 supports just about any resolution you can think of, from 64x64 up to 4 megapixels (2048x2048) - **CFG Scale:** `1` - **Steps:** They recommend 50, 20 still works but may have some quality reduction - **Sigma Shift:** Defaults to `2.02` - **Flux Guidance Scale:** Defaults to `3.5`, fiddling this value up a bit (to eg `4`) may be better - **Sampler:** Defaults to regular `Euler` - **Scheduler:** Defaults to `Flux2`, a new specialty scheduler added for Flux.2 to use, but it makes very little difference - **Prompt Images:** add up to a max of 6 images to the prompt box to be used as reference images. This uses significantly more memory. ### Flux.2 Klein - Klein is a smaller variant of Flux.2 - It is lower quality vs the full Flux.2-Dev, but runs much faster. Certain aspects of the quality can actually be better, notably visual quality seems to have been tuned better, overall intelligence is lower. - There is a 4B and a 9B variant, while the 9B is larger it often seems like the 4B is smarter. - It uses a smaller text encoder (Qwen 4B for Klein 4B, and Qwen 8B for Klein 9B). It will be autodownloaded. - Download [Klein 4b here]() or [4b base here]() - It has a distilled variant (Steps=8, CFG=1), and a "Base" variant (high steps, high CFG) - or [Klein 9b here]() or [9b base here]() (you may need to [accept a license here]()) - or [klein 9b-kv cache version here]() - This version uses "KV Cache" to accelerate image editing. - Any model with KV Cache support **MUST HAVE** `9b-kv` in the filename. This is how Swarm detects and applies KV Cache behavior to the model. (There is no way to automatically detect). - This requires significantly more VRAM, so most people do not want this. - Save the file into `diffusion_models` - Broadly works the same as Flux.2-Dev - On the distilled model set `Steps` to `8`, on base model use normal high step counts - On the distilled model set `CFG Scale` to `1`, on base model use normal CFG eg `7` - They have an [official prompting guide here]() # Chroma - Chroma is a derivative of Flux, and is supported in SwarmUI - FP8 Scaled versions here: [silveroxides/Chroma1-HD-fp8-scaled]() - Or older revs [Clybius/Chroma-fp8-scaled]() - Or GGUF versions here: [silveroxides/Chroma-GGUF]() - Or original BF16 here (not recommended): [lodestones/Chroma]() - Model files goes in `diffusion_models` - Uses standard CFG, not distilled to 1 like other Flux models - Original official reference workflow used Scheduler=`Align Your Steps` with Steps=`26` and CFG Scale=`4` - (It's named `Optimal Steps` in their workflow, but Swarm's AYS scheduler is equivalent to that) - "Sigmoid Offset" scheduler was their later recommendation, it requires a custom node - You can `git clone https://github.com/silveroxides/ComfyUI_SigmoidOffsetScheduler` into your ComfyUI `custom_nodes`, and then restart SwarmUI, and it will be available from the `Scheduler` param dropdown - Or, "power_shift" / "beta42" from [ComfyUI_PowerShiftScheduler]() may be better - Works the same, `git clone https://github.com/silveroxides/ComfyUI_PowerShiftScheduler` into your ComfyUI `custom_nodes` and restart - Generally works better with longer prompts. Adding some "prompt fluff" on the end can help clean it up. This is likely related to it being a beta model with an odd training dataset. - **Parameters** - **CFG Scale:** around `3.5` - **Sampler:** Defaults to regular `Euler` - **Scheduler:** Defaults to `Beta` - **Steps:** Normal step counts work, official recommendation is `26` - **Sigma Shift:** Defaults to `1` - **Resolution:** `1024x1024` or nearby values. The *HD* models were trained extra on `1152x1152`. # Chroma Radiance - Chroma Radiance is a pixel-space model derived from Flux, and is supported in SwarmUI - It is a work in progress, expect quality to be limited for now - Download here [lodestones/Chroma1-Radiance]() - Model files goes in `diffusion_models` - It does not use a VAE - **Parameters** - **CFG Scale:** around `3.5` - **Sampler:** Defaults to regular `Euler` - **Scheduler:** Defaults to `Beta` - **Steps:** Normal step counts work, higher is recommended to reduce quality issues - **Sigma Shift:** Defaults to `1`. Set to `0` to explicitly remove shift and use the underlying comfy default behavior. - **Prompt:** Long and detailed prompts are recommended. - **Negative Prompt:** Due to the model's experimental early train status, a good negative prompt is essential. - Official example: `This low quality greyscale unfinished sketch is inaccurate and flawed. The image is very blurred and lacks detail with excessive chromatic aberrations and artifacts. The image is overly saturated with excessive bloom. It has a toony aesthetic with bold outlines and flat colors.` # Lumina 2 ![img](/docs/images/models/lumina-2.png) *(Generated with the `highest degree of image-text alignment` preprompt, CFG=4, SigmaShift=6, Steps=20)* - Lumina 2 is an image diffusion transformer model, similar in structure to SD3/Flux/etc. rectified flow DiTs, with an LLM (Gemma 2 2B) as its input handler. - It is a 2.6B model, similar size to SDXL or SD3.5M, much smaller than Flux or SD3.5L - You can download the Comfy Org repackaged version of the model for use in SwarmUI here: - Or the `diffusion_models` variant (this version will by default load in fp8, and run a bit faster on 40xx cards) - Because of the LLM input, you have to prompt it like an LLM. - This means `a cat` yields terrible results, instead give it: `You are an assistant designed to generate superior images with the superior degree of image-text alignment based on textual prompts or user prompts. a cat` to get good results - Lumina's published reference list of prompt prefixes from [source code](https://github.com/Alpha-VLLM/Lumina-Image-2.0/blob/main/sample.py#L246): - `You are an assistant designed to generate high-quality images with the highest degree of image-text alignment based on textual prompts. ` - `You are an assistant designed to generate high-quality images based on user prompts. ` - `You are an assistant designed to generate high-quality images with highest degree of aesthetics based on user prompts. ` - `You are an assistant designed to generate superior images with the superior degree of image-text alignment based on textual prompts or user prompts. ` - `You are an assistant designed to generate four high-quality images with highest degree of aesthetics arranged in 2x2 grids based on user prompts. ` - You can absolutely make up your own though. - For longer prompts the prefix becomes less needed. - The model uses the Flux.1 VAE - **Parameters:** - **CFG**: 4 is their base recommendation - **Sigma Shift:** The default is `6` per Lumina reference script, Comfy recommends `3` for use with lower step counts, so you can safely mess with this parameter if you want to. 6 seems to be generally better for structure, while 3 is better for fine details by sacrificing structure, but may have unwanted artifacts. Raising step count reduces some artifacts. - **Steps:** The usual 20 steps is fine, but reference Lumina script uses 250(?!) by default (it has a weird sampler that is akin to Euler at 36 steps actually supposedly?) - Quick initial testing shows that raising steps high doesn't work any particularly different on this model than others, but the model at SigmaShift=6 produces some noise artifacts at regular 20 steps, raising closer to 40 cuts those out. - **Renorm CFG:** Lumina 2 reference code sets a new advanced parameter `Renorm CFG` to 1. This is available in Swarm under `Advanced Sampling`. - The practical difference is subjective and hard to predict, but enabling it seems to tend towards more fine detail # Qwen Image ![img](/docs/images/models/qwen-image.jpg) *(Qwen Image ran at CFG=4, Steps=50, Res=1328x1328. This took me about 3 minutes per image. This comparison is unfair to the other models, but this model seems intended to be a 'slow but smart' model, so this is the way to run it for now. The test prompt seems to be particularly hard on Qwen Image, I promise it's smarter than this makes it look lol.)* - [Qwen Image](https://huggingface.co/Qwen/Qwen-Image) is natively supported in SwarmUI. - Download the model here [Comfy-Org/Qwen-Image_ComfyUI]() - There's an fp8 and a bf16 variant available. The fp8 model is highly recommended. - There's the 2512 version and the original available. 2512 is newer and better. - Or, for nunchaku accelerated version that uses a bit less VRAM and runs faster, [nunchaku-tech/nunchaku-qwen-image]() - Or, other option for limited memory space, GGUF versions [city96/Qwen-Image-gguf]() - Or a distilled version here [qwen_image_distill_full_fp8_e4m3fn]() - This uses CFG=1, Steps=15 or so. - There's also a couple "Lightning" loras [lightx2v/Qwen-Image-Lightning]() for the base model, CFG=1 Steps=8 or 4 - Save it to `diffusion_models` - The text encoder is Qwen 2.5 VL 7B (LLM), and will be automatically downloaded. - It has its own VAE, and will be automatically downloaded. - SageAttention has compatibility issues, if you use Sage it will need to be disabled. - **CFG:** You can use CFG=`1` for best performance. You can also happily use higher CFGs, eg CFG=`4`, at a performance cost. - **Steps:** normal ~20 works, but higher steps (eg 50) is recommended for best quality - **Resolution:** 1328x1328 is their recommended resolution, but you can shift it around to other resolutions in a range between 928 up to 1472. - **Performance:** Can be fast on Res=928x928 CFG=1 Steps=20, but standard params are very slow (one full minute for a standard res 20 step cfg 4 image on a 4090, compared to ~10 seconds for Flux on the same). - Requires >30 gigs of system RAM just to load at all in fp8. If you have limited sysram you're gonna have a bad time. Pagefile can help. - **Prompts:** TBD, but it seems very friendly to general prompts in both natural language and booru-tag styles. Official recommendations are very long LLM-ish prompts though. - **Sigma Shift:** Comfy defaults it to `1.15`, but this ruins fine details, so Swarm defaults it to `3` instead. Many different values are potentially valid. Proper guidance on choices TBD. ### Controlnets - There are three controlnet versions available for Qwen Image currently - Regular form - There's a regular controlnet-union available here [InstantX/Qwen-Image-ControlNet-Union]() (be sure to rename the file when you save it) - works like any other controlnet. Select as controlnet model, give it an image, select a preprocessor. Probably lower the strength a bit. - Compatible with lightning loras. - If not using Lightning, probably raise your CFG a bit to ensure your prompt is stronger than the controlnet. - "Model Patch" - Download here [Comfy-Org/Qwen-Image-DiffSynth-ControlNets: model_patches]() - Save to ControlNets folder - Work the same as any other controlnets for basic usage, but advanced controls (eg start/stop steps) don't quite work - LoRA form - Download here [Comfy-Org/Qwen-Image-DiffSynth-ControlNets: loras]() - Save to loras folder - Select the lora, use with a regular qwen image base model - Upload a prompt image of controlnet input (depth or canny) - You can create this from an existing image by using the Controlnet Parameter group, select the preprocessor (Canny, or MiDAS Depth), and hit "Preview" - You cannot use the controlnet parameters directly for actual generation due to the weird lora-hack this uses - Note that Qwen Image controlnets do not work the best on the Qwen Image Edit model. ### Qwen Image Edit - The Qwen Image **Edit** model can be downloaded here: [Comfy-Org/Qwen-Image-Edit_ComfyUI]() - `qwen_image_edit_2511_fp8mixed` recommended currently - Or GGUF version here: [unsloth/Qwen-Image-Edit-2511-GGUF]() or old 2509 [QuantStack/Qwen-Image-Edit-2509-GGUF]() (or old version [QuantStack/Qwen-Image-Edit-GGUF]()) - Or nunchaku version here: [nunchaku-qwen-image-edit-2509]() (or old version [nunchaku-qwen-image-edit]()) - For original Edit or v2509, the architecture cannot be autodetected and must be set manually. 2511 can autodetect. - Click the `☰` hamburger menu on a model, then `Edit Metadata`, then change `Architecture` to `Qwen Image Edit Plus` and hit `Save` - For the original model (prior to 2509), use `Qwen Image Edit` - Most params are broadly the same as regular Qwen Image - **CFG** must be `1`, Edit is not compatible with higher CFGs normally (unless using an advanced alternate guidance option) - **Sigma Shift:** `3` or lower (as low as `0.5`) is a valid range. Some users report that a value below 1 might be ideal for single-image inputs. - You can insert image(s) to the prompt box to have it edit that image - It will focus the first image, but you can get it to pull features from additional images (with limited quality) - Qwen Image Edit Plus works with up to 3 images well - Use phrasing like `The person in Picture 1` to refer to the content of specific input images in the prompt - There are a few samples of how to prompt here - `Smart Image Prompt Resizing` parameter (top-left, under Image Prompting) will resize your input images automatically. Turn this off if you've carefully sized your images in advance. - Some versions of Qwen Edit require strict sizing to work well. 2511 reportedly works fine within a range of options. - There are a couple dedicated Qwen Image Edit Lightning Loras [lightx2v/Qwen-Image-Edit-2511-Lightning]() or for older copies [lightx2v/Qwen-Image-Lightning]() - Take care to separate the Edit lora vs the base Qwen Image lora. # Qwen Image 2.1 ![img](/docs/images/models2/qwenimage21.jpg) - [Qwen Image 2.1]() is supported in SwarmUI - This models specific claim to fame is being the first proper model that supports alpha - meaning, it can generate images with transparency - It can also do image editing & referencing at relatively high quality - They have a blog post that demonstrates main use cases here: [Qwen AI Blog: Qwen Image 2.1]() - Main model downloads are here: [Comfy-Org/Qwen-Image-2.1]() - There is a regular and turbo version - For the Turbo version, use Steps=8 and Scheduler=AlignYourSteps - It has a ch=64 scale=16 RGBA VAE, it will be downloaded automatically - Note that many models are scale=8, this is scale=16, meaning it has a larger 'snap' in resolution range, and needs to be set to 2048x2048 to generate the equivalent quality of what most models do at 1024x1024 - There is an unofficial 'texture fix' VAE that looks a bit better, download here [madebyollin/texture-fix-vae](). If you prefer it, select it in User Settings -> DefaultQwen21VAE. - Supports generating transparency (will break if image format is jpg) - **Parameters:** - **Prompt**: General VLM prompting - The preset for transparent images is `This is an RGBA image with transparency. {prompt}. The image has alpha channel and the background is transparent.` this is essentially required for proper transparent backgrounds. - **Prompt Images:** add up to 10 images to the prompt for editing or referencing - **CFG Scale:** Use `1` - **Steps:** Normal step ranges (like `20`, official Qwen reference is `40`, can still produce coherent images with lower step counts like 12) - For Turbo version, use 8 - **Sampler:** Default is fine - **Scheduler:** Default is fine - For Turbo version, AlignYourSteps is recommended - **Resolution:** The model is built for SideLength=2048 (at a 16x downscale, equivalent to 1024x1024 on most other models), but you may wish to lower this for speed (and sometimes quality, 2048 may be a bit undertrained), eg SideLength=1024 is more comfortable for quick gens - Consider trying 1024 base res and a Refiner Upscale 2x # Hunyuan Image 2.1 ![img](/docs/images/models/hy-img-21.jpg) - [Hunyuan Image 2.1](https://huggingface.co/tencent/HunyuanImage-2.1) is supported in SwarmUI. - The main model's official original download here: [tencent/HunyuanImage-2.1](https://huggingface.co/tencent/HunyuanImage-2.1/blob/main/dit/hunyuanimage2.1.safetensors), save to `diffusion_models` - FP8 download link pending - Or GGUF: [QuantStack/HunyuanImage-2.1-GGUF]() - There is also a distilled variant, you can download here: [Comfy-Org/HunyuanImage_2.1_ComfyUI](https://huggingface.co/Comfy-Org/HunyuanImage_2.1_ComfyUI/blob/main/split_files/diffusion_models/hunyuanimage2.1_distilled_bf16.safetensors). - (The tencent upload does not work, use the linked upload) - FP8 download link pending - Or GGUF: [QuantStack/HunyuanImage-2.1-Distilled-GGUF]() - They also provide and recommend a Refiner model, you can download that here: [hunyuanimage-refiner](https://huggingface.co/tencent/HunyuanImage-2.1/blob/main/dit/hunyuanimage-refiner.safetensors) - FP8 download link pending - Or GGUF: [QuantStack/HunyuanImage-2.1-Refiner-GGUF]() - This naturally is meant to be used via the Refine/Upscale parameter group in Swarm. - Set `Refiner Control Percentage` to `1`, set `Refiner Steps` to `4`, set `Refiner CFG Scale` to `1` - You may also want to mess with the prompt, official recommend is some hacky LLM stuff: `<|start_header_id|>system<|end_header_id|>Describe the image by detailing the color, shape, size, texture, quantity, text, spatial relationships of the objects and background: <|eot_id|><|start_header_id|>user<|end_header_id|> Make the image high quality<|eot_id|>`. You can use ` my prompt here that llm junk here` in Swarm to automatically emit refiner-specific prompts. - This specific model is not required. In fact, it's pretty bad. It can be replaced with other models of other architectures - pick the model with details you like and refine with that instead. - Running the base model without a refiner works too, but fine detail quality is bad. You'll want to pick a refiner. *(Possibly finetunes will fix the base in the future, as happened eg with SDXL Base years ago.)* - **CFG Scale:** Normal CFG range, recommended around `3.5`. The distilled model is capable of CFG=`1`. The refiner requires CFG=`1`. - **Steps:** Normal step values, around `20`. Refiner prefers `4`. - **Resolution:** Targets `2048x2048`, can work at lower resolutions too. - The VAE is a 32x32 downscale (vs most image models use 8x8), so it's a much smaller latent image than other models would have at this scale. - 2048 on this model is the same latent size as 512 on other models. - **Sigma Shift:** Default is `5`. Refine defaults to `4`. - TBD: Info specific to Distilled variant usage (doesn't seem to work well with their documented settings, testing TBD or comfy fix), and dedicated Refiner model # Z-Image ![img](/docs/images/models2/zimage.jpg) - [Z-Image]() and [Z-Image Turbo]() are supported in SwarmUI! - It is a 6B scaled model, with both a strong base and an official turbo designed to run extremely fast while competing at the top level of image models - "Edit" and "Omni" variants are still expected - The "Turbo" model was the first version officially released download here [Z-Image-Turbo-FP8Mixed]() - Or the original BF16 fat version [Comfy-Org/z_image_turbo]() - Or GGUF version here [jayn7/Z-Image-Turbo-GGUF]() - Save in `diffusion_models` - The "Base" model was released around 2 months later, download here (Pending: good fp8mixed) - Or the original BF16 fat version [Comfy-Org/z_image]() - Save in `diffusion_models` - Uses the Flux.1 VAE, will be downloaded and handled automatically - You might prefer swapping to the [UltraFlux VAE]() which gets better photorealism quality (be sure to rename the file when you save it, eg `Flux/UltraFlux-vae.safetensors`) - **Parameters:** - **Prompt:** Supports general prompting in any format just fine. Speaks English and Chinese deeply, understands other languages decently well too. - **Sampler:** Default is fine. Some users find `Euler Ancestral` can be better on photorealism detail. Comfy examples suggests `Res MultiStep`. - **Scheduler:** Default is fine. Some users find `Beta` can be very slightly better. - **CFG Scale:** For Turbo, `1`, for base normal CFG ranges (eg 4 or 7) - **Steps:** For Turbo, small numbers are fine. `4` will work, `8` is better. For Base, 20+ steps as normal. - Original Turbo repo suggests 5/9, but this appears redundant in Swarm. - For particularly difficult prompts, raising Steps up to `20` on Turbo or `50` on Base may help get the full detail. - **Resolution:** Side length `1024` is the standard, but anywhere up to `2048` is good. `512` noticeably loses some quality, above `2048` corrupts the image. - **Sigma Shift:** Default is `3`, raising to `6` can yield stronger coherence. - Here's a big ol' grid of Z-Image Turbo params: [Z-Image MegaGrid]() ### Z-Image Turbo Seed Variety Trick - There's a trick to get better seed variety in Z-Image Turbo: (This is less needed on Base) - Add an init image (Any image, doesn't matter much - the broad color bias of the image may be used, but that's about it). - Set Steps higher than normal (say 8 instead of 4) - Set Init Image Creativity to a relatively high value (eg 0.7) - Set Advanced Sampling -> Sigma Shift to a very high value like `22` - Hit generate. - (This basically just screws up the model in a way it can recover from, but the recovery makes it take very different paths depending on seed) ### Z-Image Controlnets - There's a "DiffSynth Model Patch" controlnet-union available here [alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1]() - This goes in your regular ControlNets folder - Comfy treats this as separate "model_patches", to use Comfy folder format, add `;model_patches` to the end of Server Config->Paths->SDControlNetsFolder - Proper Architecture ID is `Z-Image ControlNet (DiffPatch)` - Works like any other controlnet. Select as controlnet model, give it an image, select a preprocessor. Fiddle the strength to taste. - Despite being a Union controlnet, the Union Type parameter is not used. - Because it is "Model Patch" based, the Start and End parameters also do not work. # Kandinsky 5 - Kandinsky 5 Image Lite is supported in SwarmUI - Also the video models, docs [in the video model support doc](/docs/Video%20Model%20Support.md#kandinsky-5) - There are multiple variants, pick one to download from here: [kandinskylab/kandinsky-50-image-lite]() - **Parameters:** - **CFG Scale:** regular CFG such as `5` works. - **Steps:** Regular 20+ steps. - **Resolution:** Side length 1024. # Anima - [Anima by Circlestone Labs]() is a 2B anime model build on Cosmos, and it is fully supported in SwarmUI. - It is designed to be tiny, lightweight, fast, but built on a strong architecture. - It is the first model architecture publicly released that was sponsored by Comfy Org! - It is explicitly still in Preview status, they will be training it further before it's entirely ready. - Download [Anima v1 here]() - Save in `diffusion_models` - You can also get [Anima Controlnets here]() - Pick your favorites (depth, pose, lineart, scribble are the main ones) - Save to the `controlnet` model folder - Use like any other controlnet - It uses a tiny Qwen 3 600M ("0.6B") text encoder. This will be autodownloaded. - It uses the Qwen Image VAE. This will be autodownloaded. - **Parameters:** - **Prompt:** Trained on both booru style tag prompts (`1girl, etc`) and natural language prompts. They have official specific writing guidance [here]() - **CFG Scale:** Regular CFG scales (eg `4`) work. - **Steps:** Regular 20+ steps. - **Resolution:** Side length 1024 recommend, but any lower value works too. Higher values do not work well. Refiner upscale needs tiling due to corruption at high res. - **Sampler:** Defaults to `ER-SDE-Solver`, but all common samplers work. They officially recommend also trying out `Euler Ancestral` or `DPM++ 2M SDE` - **Scheduler:** Default is fine (`Simple`), or you can experiment at will. The model is adaptable. # HiDream-O1 ![img](/docs/images/models2/hidreamo1.jpg) *(HiDream-o1 Dev FP8, Steps=20, CFG=1, SideLength=2048)* - HiDream's [HiDream O1]() is supported in SwarmUI! - It is an 8B model, with both a base and an official 'dev' distill designed to run faster - The "Dev" model (in fat BF16) can be downloaded here [Comfy-Org/HiDream-O1-Image - dev]() - Dev FP8 version can be downloaded here [Comfy-Org/HiDream-O1-Image - dev FP8]() - Or the base version (in fat BF16) [Comfy-Org/HiDream-O1-Image - base]() - Base FP8 version can be downloaded here [Comfy-Org/HiDream-O1-Image - base FP8]() - Save in `Stable-Diffusion` - It has no VAE, but has in-middle dedicated large patch scaling to compensate - Its text encoding is similarly native-integrated - **Parameters:** - **Prompt:** Supports general prompting in any format just fine. Speaks at least English and Chinese. Was designed to use LLM-written prompts. - **Prompt Images:** You can upload up to 10 images for image-editing input, but it only strongly obeys single-image input. - **Sampler:** Default is fine. - **Scheduler:** Default is fine. - **CFG Scale:** For Dev, `1`, for base normal CFG ranges (around `5`) - **Steps:** For Dev `28` is their recommendation, but even just `4` works fine for simple images. For Base, 50 steps is the official recommendation. - **Resolution:** Side length `2048` is the model's standard, but a wide range works well. - Because of the aggressive patch scaling, 2048 on this model looks more like 1024 on most other models. 1024 on this model looks noticeably worse. Going above 2048 will have some color distortion. - **Dev Lora:** - A dev lora can be downloaded here [Kijai/hidream-O1-image_comfy](). It allows use of the base model with the distilled behavior from the Dev model. 8 steps will generate a coherent image of lower quality, 16 steps seems closer to original quality. Use CFG Scale 1. # Ideogram 4 - [Ideogram 4]() is supported in SwarmUI! - It is a 9B model with an optional split unconditional model - You can download the FP8 here: [Comfy-Org/Ideogram-4 FP8]() - Or the NVFP4 (5 gigs) here: [Comfy-Org/Ideogram-4 nvfp4]() - Or the int8 here [silveroxides/ideogram4-int8]() - You will also want the "Unconditional" model [Comfy-Org/Ideogram-4 uncond FP8]() - Or the NVFP4 [Comfy-Org/Ideogram-4 uncond nvfp4]() - Or the int8 here [silveroxides/ideogram4-int8-uncond]() - The idea is you use a separate model for the negative half of CFG from the positive half - this is not required, but it improves quality - Select it under Advanced Model Addons -> Negative Model - It has built-in-to-the-model censorship, the model itself will try to reject inappropriate prompts. - **Parameters:** - **Prompt:** They have an official prompting guide here [Ideogram-OSS: Docs/Prompting]() - They suggest long form JSON prompts, and have trained the model to understand features within such as bounding box coordinates as part of the structure - If you don't use JSON it will just censor you almost every time. - **Prompt Images:** Ideogram can (sorta) take reference images in the prompt. Improvements seemingly TBD. - **Steps:** They suggest `12` for Turbo, `48` for quality. Anywhere in between is fine. - **CFG:** Standard range around `7`, they suggest using Refiner with RefinerMethod=StepSwap, Control Percentage low (1-3 steps), and RefinerCFG=3. - **Sampler:** Default is fine. - **Scheduler:** Default is `Ideogram4 Default`, an official custom/unique scheduler. There is also a `Turbo` variant which may work better for low step count gens. Some users have been using other schedulers (eg `Normal`, `Flux.2`, etc.) with better results than the official scheduler. - **Resolution:** Side length `1024` is the default. - **Sigma Shift:** Default is `5`, but `1` is the legacy default. Sigma shift does nothing on the Ideogram official scheduler but does do things on normal schedulers. # Krea 2 ![img](/docs/images/models2/krea2.jpg) - Krea 2 is supported in SwarmUI! - It is a 13B model (reported officially as 12B) with a Base ('Raw') and Turbo variant. - Downloads of the model are available at [Comfy-Org/Krea-2]() - Or an int8 from [silveroxides/K2Q-turbo-int8]() - Pick one of the 'turbo' for normal generation, or there's also a 'raw' (base) model that's intended for training usage but may be useful for some inference experiments - Pick your preferred size (fp8 recommended, nv4 for memory-limited computers, bf16 for research usage) - Save your choice(s) in `diffusion_models` - There is also an official raw-to-turbo Lora available [Comfy-Org/Krea-2/Loras]() - It has built-in censorship, the model will not generate risque things with common prompts (but can be tricked or bypassed). - The most common baseline bypass is [Krea2FilterBypass LoRA](), but almost any nsfw lora will naturally bypass the filter. - Uses Qwen 3 VL 4B as a text encoder, and the QwenImage VAE, these will be automatically downloaded. - **Parameters:** - **Prompt:** Normal general model prompting works as expected, but NSFW terms will be stripped by the models internal text-refiner. - **Prompt Images:** Krea 2 is not an editing model per se, but it can take in reference images in the prompt (by default only the text encoder sees this, not the core model). - There are unofficial "Ostris" edit loras like this: [Ostris/Krea2 Style Reference](), results may vary. - For those, set **Enable Reference Latents** under Image Prompting to **Index Timestep Zero (Ostris)**. - There are also "identity edit" LoRAs like this: [Krea Identity Edit]() - use **Index (Identity Edit)** on that same dropdown. - **Resolution:** Side length `1024` is the default, but it work anywhere from `128` to `4096`. - **Steps:** For Turbo, `8` recommended, `4` minimum. For Base normal step counts (20+). - **CFG:** For Turbo, `1`. For Base normal CFG ranges (4+? 7? idk, tbd) - **Sampler:** Default is fine. - **Scheduler:** Default is fine. - **Sigma Shift:** Defaults to `1.15` # Boogu ![img](/docs/images/models2/boogu.jpg) - [Boogu]() is supported in SwarmUI! - It is a 10B model, derived from OmniGen, that comes in three separate variants - **Base**, **Edit**, and **Turbo**: - **Base**: [Comfy-Org/Boogu-Image - Base FP8]() - **Turbo** - very fast, low step counts: [Comfy-Org/Boogu-Image - Turbo FP8]() - Or the NVFP4 (6 gigs) here: [Comfy-Org/Boogu-Image - Turbo nvfp4]() - **Edit** - image editing, takes a prompt image as a reference: [Comfy-Org/Boogu-Image - Edit FP8]() - Or the NVFP4 (6 gigs) here: [Comfy-Org/Boogu-Image - Edit nvfp4]() - **Edit Turbo** - image editing, takes a prompt image as a reference, turbo step/cfg handling: [Comfy-Org/Boogu-Image - Edit Turbo int8]() - Uses the Flux.1 VAE and a Qwen3-VL8B text encoder, both downloaded and handled automatically. - For editing, use the `Edit` model and give it a prompt image - SwarmUI automatically wires it in as the reference. - **Parameters:** recommendations differ between the Base, Edit, and Turbo models: - **Sampler:** For Turbo, use LCM, otherwise default is fine, DPM++ 2M is recommended as a bit better than Euler for these models. - **Scheduler:** For Turbo, use SGM Uniform, otherwise default is fine. - **CFG Scale:** For Turbo, `1`, otherwise normal CFG ranges (around `4`). - **Steps:** For Turbo, `4` is recommended, otherwise `20` as normal. - **Resolution:** Side length `1024` is the default. - **Sigma Shift:** Default is `3.16`. # Ming Image - [Ming Image]() is supported in SwarmUI! - It is a 6B model intended for infographic generation - Download here: [Comfy-Org/Ming-Image]() - There is also a 'layer' model, not currently supported - Supports generating transparency (will break if image format is jpg) - VAE and text encoder will be automatically downloaded - **Parameters:** - **Prompt:** Likes JSON blocks, similar to ideogram - **CFG Scale:** `1` - **Steps:** They recommend `12` # Flux 3 Image ![img](/docs/images/models2/flux3klein.jpg) *(Flux 3 Klein SD, 16 steps, 1024)* - [FLUX 3 Image Klein]() is supported in SwarmUI! - It is a 5.6B image model intended as a direct partial upgrade to the Flux 2 Klein models - Download here: [Black-Forest-Labs/Flux-3-Image-Klein]() - They have a "base" and "sd" version - "sd" means distilled (CFG=1, Steps=8 or 16) - They have bf16, int8, and nvfp4 versions - the int8 version is strongly recommended for most users - Initial community opinion is low regarding quality as compared to current leading image models, but it is a very *fast* model at least - Supports text-to-image, and image editing with prompt image references - VAE and text encoder will be automatically downloaded - The VAE is still the Flux 2 image VAE - **Parameters:** - **Prompt:** They recommend using an LLM to rewrite your prompts to be much longer - They allow JSON bounding box prompt in this example format: `Minimalist graphic illustration featuring a black silhouette of a person centered against a solid, vibrant chartreuse background . The figure is captured in a dynamic, mid-stride running pose, facing toward the left side of the frame. [{"id": "background_1", "bbox": [0, 0, 1000, 1000], "desc": "A flat field of neon yellow-green with a subtle paper texture and fine, organic grain."}, {"id": "silhouette_1", "bbox": [150, 150, 850, 850], "desc": "A black silhouette of a person in motion, made of dense stippled ink."}]` - They have LLM system prompts available here [Black-Forest-Labs/system_prompts]() - Also, prompting guide [bfl.ai]() - **Prompt Images:** You can add up to 10 images to the prompt to use as references or editing - **CFG Scale:** `1` for 'sd' Distilled, regular ranges (eg `3.5`) for Base - **Steps:** `16` for 'sd' Distilled (can go down to `4` at some quality loss), regular ranges (`20` or more) for Base - **Resolution:** `1024` is default but anywhere from `128` up to `3072` can work - **Sampler:** Defaults to `er_sde` # MOSS Image ![img](/docs/images/models2/mossimage.jpg) - [MOSS Image]() is supported in SwarmUI! - It is a 5B image model - Download here [Comfy-Org/MOSS-Image]() - There are bf16 and int8 variants, the int8 is recommended - There is a 'Base' and 'SFT' version - the 'SFT' is quality tuned, whereas the base is less trained and more variable - The VAE (Flux 2) and Text Encoder (Qwen 3.5 9B) will be automatically downloaded - **Parameters:** - **Prompt:** They have an official prompt enhancer LLM skill file [OpenMOSS/prompt-enhancer]() - Supports JSON Bounding Boxes in captions, see [their example JSON prompt]() - **Prompt Images:** The model does not natively do image editing or references, but you can toss in prompt images anyway and the textencoder will see them, just not retain all details - **CFG Scale:** normal ranges, they officially use `3` but higher may be better - `7` is used for the sample image - **Steps:** normal ranges like `20` though higher may be needed for the raw models - **Resolution:** `1024`, lower values down to 256 work as expected, higher values are quite variable - **SkipLayerGuidance:** The OpenMOSS team recommends skip layer guidance with Scale set to `0` and Layer Target set to `22`, but the benefit is questionable. This is available in Swarm's extension list if you'd like to try it. # Bria Fibo 2 ![img](/docs/images/models2/fibo2.jpg) - [Fibo 2]() is supported in SwarmUI! - It is an 8B model with a "dev" model and an Edit model - Download here [Comfy-Org/Fibo-2.0]() - int8_convrot recommended - You can probably ignore dev and just grab the Edit, it does regular t2i fine - Save in `diffusion_models` - The VAE (Flux 2) and Text Encoder (Qwen 3 VL 4B) will be automatically downloaded - **Parameters:** - **Prompt:** This model was trained for very long JSON prompts. - It works without those, but the quality and understanding is poor. - **CFG Scale:** `1` - **Steps:** `8` or higher for dev, `4` for edit - **Resolution:** defaults to `1024` but can do anywhere from `256` to `1536` well, starts losing coherence above that but not unusuable # MiniMax H3 Image ![img](/docs/images/models2/h3image.jpg) - [MiniMax H3 Image](https://huggingface.co/MiniMaxAI/MiniMax-H3-Image) is supported in SwarmUI! - It is a 20B direct text2image that supports references (including for editing) - Download here [Comfy-Org/MiniMax-H3-Image]() - Save in `diffusion_models` - It uses Qwen3 VL 32B as a text encoder, Swarm will automatically download an fp4 (16GB) version of it. - It has a unique image VAE. - **Parameters:** - **Prompt:** Supports H3 syntax like ``, ``, etc. - **Prompt Images:** You can add up to 16 images to the prompt to use as references or editing - **CFG Scale:** `1` - **Steps:** `6` - **Resolution:** Defaults to `2048` but can do from `768`-ish (will degrade heavily below that) up to `4096`. - It is a 16x downscale VAE, vs most other models are 8x, so this 2048 is closer to other models' 1024. - The best strat is probably gen at 4096, then use SeedVR with a pre-downscale and re-upscale to fix the details. # Video Models - Video models are documented in [Video Model Support](/docs/Video%20Model%20Support.md). - You can use some (not all) Text2Video models as Text2Image models. - Generally, just set **Text2Video Frames** to `1` and it will be treated as image gen. - Some models may favor different parameters (CFG, Steps, Shift, etc.) for images vs videos. # Alternative Model Formats ## Bits-and-Bytes NF4 Format Models - BnB NF4 and FP4 format models, such as this copy of Flux Dev [lllyasviel/flux1-dev-bnb-nf4](), are partially supported in SwarmUI automatically. - The detection internally works by looking for `bitsandbytes__nf4` or `bitsandbytes__fp4` in the model's keys - The first time you try to load a BNB-NF4 or BNB-FP4 model, it will give you a popup asking to install support - This will autoinstall [silveroxides/ComfyUI_bnb_nf4_fp4_Loaders]() which is developed by silveroxides, comfyanonymous, and lllyasviel, and is under the AGPL license. - You can accept this popup, and it will install and reload the backend - Then try to generate again, and it should work - Note that BnB-NF4 and BNB-FP4 models have multiple compatibility limitations, including even LoRAs don't apply properly. - If you want a quantized flux model, GGUF is recommended instead. - Support is barely tested, latest bnb doesn't work with comfy but old bnb is incompatible with other dependencies, good luck getting it to load. - Seriously, just use GGUF or something. bnb is not worth it. ## GGUF Quantized Models - GGUF Quantized `diffusion_models` models are supported in SwarmUI automatically. - The detection is based on file extension. - They go in `(Swarm)/Models/diffusion_models` and work similar to other `diffusion_models` format models - Required VAE & TextEncoders will be autodownloaded if you do not already have them. - The first time you try to load a GGUF model, it will give you a popup asking to install support - This will autoinstall [city96/ComfyUI-GGUF]() which is developed by city96. - You can accept this popup, and it will install and reload the backend - Then try to generate again, and it should just work ## Nunchaku (MIT Han Lab) - MIT Han Lab's "[Nunchaku]()" / 4-bit SVDQuant models are a unusual quant format that is supported in SwarmUI. - Nunchaku is a very dense quantization of models (eg 6GiB for Flux models) that runs very fast (4.4 seconds for a 20 step Flux Dev image on Windows RTX 4090, vs fp8 is ~11 seconds on the same) - It is optimized for modern nvidia GPUs, with different optimizations per gpu generation - RTX 30xx and 40xx cards need "int4" format nunchaku models - RTX 50xx or newer cards need "fp4" format nunchaku models - They go in `(Swarm)/Models/diffusion_models` and work similar to other `diffusion_models` format models - Make sure you download a "singlefile" nunchaku file, not a legacy "SVDQuant" folder - Required VAE & TextEncoders will be autodownloaded if you do not already have them. - For the older "SVDQuant" Folder Models [mit-han-lab/svdquant](), The detection is based on the folder structure, you need the files `transformer_blocks.safetensors` and `comfy_config.json` inside the folder. You cannot have unrelated files in the folder. - The first time you try to load a Nunchaku model, it will give you a popup asking to install support - This will autoinstall [mit-han-lab/ComfyUI-nunchaku]() and its dependencies - You can accept this popup, and it will install and reload the backend - Then try to generate again, and it should just work - Nunchaku has various compatibility limitations due to hacks in the custom nodes. Not all lora, textenc, etc. features will work as intended. - It does not work on all python/torch/etc. versions, as they have deeply cursed dependency distribution - The `Nunchaku Cache Threshold` param is available to enable block-caching, which improves performance further at the cost of quality. ## TensorRT - TensorRT support (`.engine`) is available for SDv1, SDv2-768-v, SDXL Base, SDXL Refiner, SVD, SD3-Medium - TensorRT is an nvidia-specific accelerator library that provides faster SD image generation at the cost of reduced flexibility. Generally this is best for heavy usages, especially for API/Bots/etc. and less useful for regular individual usage. - You can generate TensorRT engines from the model menu. This includes a button on-page to autoinstall TRT support your first time using it, and configuration of graph size limits and optimal scales. (TensorRT works fastest when you generate at the selected optimal resolution, and slightly less fast at any dynamic resolution outside the optimal setting.) - Note that TensorRT is not compatible with LoRAs, ControlNets, etc. - Note that you need to make a fresh TRT engine for any different model you want to use. # Obscure Model Redirection ### Stable Diffusion v1 and v2 ### SegMind SSD-1B ### Stable Cascade ### PixArt Sigma ### AuraFlow ### NVIDIA Sana ### HiDream-i1 ### Cosmos Predict2 ### OmniGen 2 ### Ovis # ERNIE # Lens These obscure/old/bad/unpopular/etc. models have been moved to [Obscure Model Support](/docs/Obscure%20Model%20Support.md)