Model guides
Run image and video models locally.
One guide per model family: the files ComfyUI needs, the folders they go in, what fits your graphics card or Mac, and the settings that work. Every family here also runs built in with HEISS UI.
Which one should I run?
A short answer for the common cases. Each guide has the details and the alternatives.
- An 8 GB graphics card and good pictures fastFlux.2 Klein
- The most LoRAs, styles and fine-tunes, on 6 to 8 GBSDXL
- Photographic images on 16 GB or moreKrea 2
- Legible text in the picture: signs, posters, labelsQwen-Image
- Anime and illustrationAnima
- Video, on 16 GB or moreWan 2.2
Image models.
Start with these. The first three are the models HEISS UI offers on an empty studio.
More image models.
Larger, newer or more specialised. Check each licence: several are non-commercial.
Anime.
Models trained for anime and illustration. SDXL’s Illustrious and Pony lines are on the SDXL page.
Stable Diffusion, and Sana.
The Stable Diffusion models the rest grew from, and NVIDIA’s small, fast Sana.
Video models.
Beta in HEISS UI. Plan on 16 GB or more, and plenty of system memory.
By graphics card.
What runs on your GPU or Mac, and how to set it up.
Questions.
Which model should I start with?
On an 8 GB graphics card, Flux.2 Klein 4B in fp8 or an SDXL checkpoint. On 16 GB, Krea 2 Turbo, Z-Image Turbo or Flux.2 Klein 9B. On 24 GB, Qwen-Image in fp8 fits, and Flux.2 Dev runs in fp8 with 64 GB of system memory. On a Mac, count the unified memory: 16 GB runs SDXL, 18 GB Flux.2 Klein 4B.
What do I need to run these models locally?
ComfyUI, a graphics card or a Mac with Apple Silicon, and the model’s files: usually a diffusion model, one or more text encoders and a VAE. Each guide lists them with their sizes and the ComfyUI folder they go in.
What’s the difference between fp16, fp8 and GGUF files?
They are the same model stored at different precisions. fp16 or bf16 is the original, fp8 is about half the size with a small quality cost, and GGUF files (Q8_0 down to Q4) are smaller still and load through the ComfyUI-GGUF nodes. On a Mac, fp8 saves disk space but not memory, since it loads at full precision.
Do these models run on a Mac?
Most image models do, in ComfyUI on Apple Silicon, slower than on a comparable NVIDIA card. Pick full-precision or GGUF files rather than fp8. Video models are slow on a Mac and some samplers don’t work there; each guide says what to expect.
Can I use these models commercially?
It depends on the model. SDXL, Flux.2 Klein 4B, Z-Image, Chroma and Wan are open for commercial use. Flux.1 Dev, Flux.2 Dev, Klein 9B and Ideogram 4 are non-commercial, Krea 2 has a revenue limit, and HunyuanVideo 1.5 and MiniMax H3 exclude some regions. Each guide’s spec card names the licence.
HEISS UI
All of them, without the wiring.
HEISS UI is a free studio on top of the ComfyUI you already have. Drop in any model on this page and it just runs.
- Drop in the file and it runs. It knows the model from the file itself, even renamed, and picks settings that work.
- Missing parts, shown first. Each one listed with its size and a button. Get all checks free space, and downloads resume and are verified.
- A first model that fits. An empty studio offers Krea 2, Flux.2 and SDXL, with the version for your GPU or Mac marked.
- A phone studio. Prompt, browse and share from the couch while the computer renders.
- Hidden. A locked, encrypted place for private images, opened with Touch ID, Windows Hello or a password.
Free and open source. macOS, Windows and Linux. Runs on your ComfyUI.
