Supported models and features

Choose a registered model below, then start with the runner. These tables describe this source checkout; installed releases can differ. ✓ means the adapter declares support; — means it does not. Support still depends on the checkpoint, inputs, hardware, kernels, and flag combinations. Follow the linked guides for restrictions; a row does not promise every feature can be enabled together. CUDA, ROCm, and Ascend availability varies by backend. INT8 names the platform instead, because its backends differ: torchao on CUDA, AITER on ROCm gfx950.

Parallelism

Ulysses and Ring split sequence/attention work; CFG splits guidance branches; PipeFusion splits transformer stages. TP is transformer tensor parallelism. FSDP and text-encoder TP reuse existing ranks. VAE D means decode, E means encode support. No current runner model supports transformer TP (--tensor_parallel_degree > 1). For the legacy FLUX feed-forward TP path, see legacy TP examples. Text-encoder TP is separate; see the Text TP column.

Model Ulysses Ring CFG PipeFusion DP TP FSDP Text TP VAE
CausalWan — — — — ✓ — ✓ — D+E
Chroma1-HD ✓ — ✓ — ✓ — ✓ — —
Cosmos3-Nano ✓ ✓ ✓ — ✓ — ✓ — D+E
Cosmos3-Super ✓ ✓ ✓ — ✓ — ✓ — D+E
FastH3 ✓ — — — — — ✓ ✓ D
FastH3-Dense ✓ — — — — — ✓ ✓ D
FastVideo/FastVideo-FastH3-8-Step-V2 ✓ — — — — — ✓ ✓ D
flux-3-image-klein ✓ ✓ — — ✓ — ✓ — D
flux-3-image-klein-base ✓ ✓ ✓ — ✓ — ✓ — D
FLUX.1-dev ✓ ✓ — ✓ ✓ — ✓ — D
FLUX.1-Kontext-dev ✓ ✓ — — ✓ — ✓ — D+E
FLUX.2-dev ✓ ✓ — ✓ ✓ — ✓ — D+E
FLUX.2-klein-4B ✓ ✓ — ✓ ✓ — ✓ — D+E
FLUX.2-klein-9B ✓ ✓ — ✓ ✓ — ✓ — D+E
HunyuanVideo ✓ ✓ — — ✓ — ✓ — D
Hunyuanvideo-1.5 ✓ ✓ — — ✓ — — — D+E
Hunyuanvideo-1.5-Distilled ✓ ✓ — — ✓ — — — D+E
Hunyuanvideo-1.5-Sparse ✓ ✓ — — ✓ — — — D+E
Ideogram-4 ✓ ✓ ✓ — ✓ — ✓ — D
Krea-2-Raw ✓ — — — ✓ — ✓ — D
Krea-2-Turbo ✓ — — — ✓ — ✓ — D
LingBot-Video-Dense ✓ — ✓ — ✓ — ✓ — D+E
LingBot-Video-MoE ✓ — ✓ — ✓ — ✓ — D+E
LTX-2 ✓ ✓ — — ✓ — ✓ — D
LTX-2.3 ✓ ✓ — — ✓ — ✓ — D
LTX-2.5 ✓ ✓ — — ✓ — — — D
LTX-2.5-full ✓ ✓ — — ✓ — — — D
LTX-Video-0.9.7-dev ✓ ✓ ✓ — ✓ — ✓ — —
Lumina2 ✓ — ✓ — ✓ — — — —
MiniMax-H3 ✓ — — — — — ✓ ✓ D
MiniMax-H3-Image ✓ — — — — — ✓ ✓ D
MiniMax-H3-Ref2VA ✓ — — — — — ✓ ✓ D
MiniMax-H3-Turbo ✓ — — — — — ✓ ✓ D
MOSS-Image ✓ ✓ ✓ — ✓ — — — —
Prism ✓ — — — — — — — —
Qwen-Image ✓ ✓ — ✓ ✓ — ✓ — D
Qwen-Image-2.1 ✓ — — — ✓ — ✓ — —
Qwen-Image-2.1-Turbo ✓ — — — ✓ — ✓ — —
Qwen-Image-Edit ✓ ✓ ✓ ✓ ✓ — ✓ — D+E
SD3.5 ✓ ✓ ✓ ✓ ✓ — ✓ — D
SkyReels-V2-I2V-14B ✓ ✓ — — ✓ — ✓ — —
SkyReels-V2-T2V-14B ✓ ✓ — — ✓ — ✓ — —
VDN-H3 ✓ — — — — — — ✓ D
Wan2.1-I2V ✓ ✓ ✓ — ✓ — ✓ — D+E
Wan2.1-T2V ✓ ✓ — — ✓ — ✓ — D
Wan2.1-T2V-1.3B ✓ ✓ — — ✓ — ✓ — D
Wan2.1-VACE-14B ✓ ✓ — — ✓ — ✓ — D+E
Wan2.2-Distilled-I2V ✓ ✓ — — ✓ — ✓ — D+E
Wan2.2-I2V ✓ ✓ ✓ — ✓ — ✓ — D+E
Wan2.2-T2V ✓ ✓ — — ✓ — ✓ — D
Wan2.2-TI2V ✓ ✓ ✓ — ✓ — ✓ — D+E
Z-Image ✓ ✓ ✓ — ✓ — ✓ — D
Z-Image-Turbo ✓ ✓ — — ✓ — ✓ — D

Memory and precision

GEMM formats apply to declared linear-layer targets, not the entire model. Text FP8 is a separate opt-in for supported encoders. VAE tiling and slicing are independent memory controls.

Model FP8 FP4 FP6 A6W4 INT8 Text FP8 Tiling Slicing
CausalWan — — — — — — ✓ ✓
Chroma1-HD — — — — — — — —
Cosmos3-Nano ✓ ✓ — — — — ✓ ✓
Cosmos3-Super ✓ ✓ — — — — ✓ ✓
FastH3 ✓ ✓ — — — — — —
FastH3-Dense ✓ ✓ — — — — — —
FastVideo/FastVideo-FastH3-8-Step-V2 ✓ ✓ — — — — — —
flux-3-image-klein ✓ — — — — — ✓ ✓
flux-3-image-klein-base ✓ — — — — — ✓ ✓
FLUX.1-dev ✓ — — — — ✓ ✓ ✓
FLUX.1-Kontext-dev ✓ — — — — ✓ ✓ ✓
FLUX.2-dev ✓ ✓ — — — ✓ ✓ ✓
FLUX.2-klein-4B ✓ — — — — ✓ ✓ ✓
FLUX.2-klein-9B ✓ — — — — ✓ ✓ ✓
HunyuanVideo ✓ — — — — ✓ ✓ ✓
Hunyuanvideo-1.5 ✓ — — — — — ✓ ✓
Hunyuanvideo-1.5-Distilled ✓ — — — — — ✓ ✓
Hunyuanvideo-1.5-Sparse — — — — — — ✓ ✓
Ideogram-4 ✓ ✓ — — — — ✓ ✓
Krea-2-Raw ✓ ✓ — — — — ✓ ✓
Krea-2-Turbo ✓ ✓ — — — — ✓ ✓
LingBot-Video-Dense ✓ ✓ — — — — ✓ ✓
LingBot-Video-MoE ✓ ✓ — — — — ✓ ✓
LTX-2 ✓ — — — — — ✓ ✓
LTX-2.3 — — — — — — ✓ ✓
LTX-2.5 ✓ ✓ — — — — ✓ —
LTX-2.5-full ✓ ✓ — — — — ✓ —
LTX-Video-0.9.7-dev — — — — — — ✓ ✓
Lumina2 — — — — — — ✓ ✓
MiniMax-H3 ✓ ✓ — — — — — —
MiniMax-H3-Image ✓ — — — — — — —
MiniMax-H3-Ref2VA ✓ ✓ — — — — — —
MiniMax-H3-Turbo ✓ ✓ — — — — — —
MOSS-Image — — — — — — ✓ ✓
Prism — — — — — — ✓ —
Qwen-Image ✓ — — — — ✓ ✓ ✓
Qwen-Image-2.1 ✓ — — — — — ✓ ✓
Qwen-Image-2.1-Turbo ✓ — — — — — ✓ ✓
Qwen-Image-Edit ✓ — — — — ✓ ✓ ✓
SD3.5 ✓ — — — — ✓ ✓ ✓
SkyReels-V2-I2V-14B — — — — — — ✓ ✓
SkyReels-V2-T2V-14B — — — — — — ✓ ✓
VDN-H3 ✓ — — — — — — —
Wan2.1-I2V ✓ ✓ ✓ ✓ — ✓ ✓ ✓
Wan2.1-T2V ✓ ✓ ✓ ✓ — — ✓ ✓
Wan2.1-T2V-1.3B ✓ ✓ ✓ ✓ — — ✓ ✓
Wan2.1-VACE-14B ✓ — — — — ✓ ✓ ✓
Wan2.2-Distilled-I2V ✓ ✓ ✓ — — ✓ ✓ ✓
Wan2.2-I2V ✓ ✓ ✓ ✓ — ✓ ✓ ✓
Wan2.2-T2V ✓ ✓ ✓ ✓ ROCm gfx950 — ✓ ✓
Wan2.2-TI2V ✓ ✓ ✓ ✓ — ✓ ✓ ✓
Z-Image ✓ — — — CUDA ✓ ✓ ✓
Z-Image-Turbo ✓ — — — CUDA ✓ ✓ ✓

Communication, caching, and adapters

FP8 communication and Attention A2A optimize Ulysses exchanges. Sparse means at least one sparse attention family is declared; consult attention backends for model/kernel restrictions. Cache methods come from the model presets; pipefusion additionally requires a compatible DBCache layout and is described in the PipeFusion guide. LoRA means supported startup fusion, not runtime adapter switching.

Model FP8 comms Attention A2A Sparse Cache presets LoRA
CausalWan — — — dbcache —
Chroma1-HD — — — — —
Cosmos3-Nano — — — — —
Cosmos3-Super — — — — —
FastH3 — — ✓ — —
FastH3-Dense — — ✓ — —
FastVideo/FastVideo-FastH3-8-Step-V2 — — ✓ — —
flux-3-image-klein ✓ — ✓ — —
flux-3-image-klein-base ✓ — ✓ — —
FLUX.1-dev ✓ — — dbcache, teacache ✓
FLUX.1-Kontext-dev ✓ — — dbcache —
FLUX.2-dev ✓ — — dbcache, fbcache —
FLUX.2-klein-4B ✓ — — fbcache —
FLUX.2-klein-9B ✓ — — fbcache —
HunyuanVideo — — — dbcache —
Hunyuanvideo-1.5 — — — dbcache —
Hunyuanvideo-1.5-Distilled — — — dbcache —
Hunyuanvideo-1.5-Sparse — — ✓ dbcache —
Ideogram-4 — — — — —
Krea-2-Raw — — — — —
Krea-2-Turbo — — — — —
LingBot-Video-Dense — — — — —
LingBot-Video-MoE — — — — —
LTX-2 — — — dbcache —
LTX-2.3 — — — dbcache —
LTX-2.5 — — — — —
LTX-2.5-full — — — — —
LTX-Video-0.9.7-dev — — — — —
Lumina2 — — — — —
MiniMax-H3 — — ✓ — —
MiniMax-H3-Image — — — — —
MiniMax-H3-Ref2VA — — ✓ — —
MiniMax-H3-Turbo — — ✓ — ✓
MOSS-Image — — — — —
Prism — — ✓ — —
Qwen-Image ✓ — — dbcache —
Qwen-Image-2.1 — — — — —
Qwen-Image-2.1-Turbo — — — — —
Qwen-Image-Edit ✓ — — dbcache —
SD3.5 ✓ — — dbcache —
SkyReels-V2-I2V-14B — — — — —
SkyReels-V2-T2V-14B — — — — —
VDN-H3 — — — — —
Wan2.1-I2V — ✓ ✓ dbcache —
Wan2.1-T2V — — ✓ dbcache —
Wan2.1-T2V-1.3B — — ✓ dbcache —
Wan2.1-VACE-14B — — — dbcache —
Wan2.2-Distilled-I2V — — ✓ dbcache —
Wan2.2-I2V ✓ ✓ ✓ dbcache, fbcache —
Wan2.2-T2V ✓ ✓ ✓ dbcache, fbcache —
Wan2.2-TI2V ✓ — ✓ dbcache, fbcache —
Z-Image ✓ — — dbcache —
Z-Image-Turbo ✓ — — — —

Accepted model names and requirements

Aliases in the same row use the same adapter. Checkpoint-specific task, precision, and resolution choices can still differ; the source link shows those rules. source means Diffusers must be installed from source. The version column records each adapter's declared minimum; an undeclared floor means the minimum is unknown. It is not a complete environment lock. See installation and containers. Input lists prompt text and conditioning media accepted by each adapter. Optional and task-specific inputs are labeled; check the source for image counts and other restrictions.

Model / source Accepted --model names Input Output Diffusers minimum
CausalWan CausalWan Text; i2v requires one image video 0.35.2
Chroma1-HD Chroma1-HD, lodestones/Chroma1-HD Text image 0.35.0
Cosmos3-Nano Cosmos3-Nano, nvidia/Cosmos3-Nano Text; optional image video 0.39.0
Cosmos3-Super Cosmos3-Super, nvidia/Cosmos3-Super Text; optional image video 0.39.0
FastH3 FastH3, FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree, FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-Synthetic-Step1300, FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-Synthetic-Step1900 Text (t2va) video + audio 0.40.0
FastH3-Dense FastH3-Dense, FastVideo/FastVideo-FastH3-4-step-Preview-v1-Dense-DataFree Text (t2va) video + audio 0.40.0
FastVideo/FastVideo-FastH3-8-Step-V2 FastVideo/FastVideo-FastH3-8-Step-V2 Text (t2va) video + audio 0.40.0
flux-3-image-klein flux-3-image-klein, FLUX.3-image-klein, black-forest-labs/flux-3-image-klein Text; optional images image source
flux-3-image-klein-base flux-3-image-klein-base, FLUX.3-image-klein-base, black-forest-labs/flux-3-image-klein-base Text; optional images image source
FLUX.1-dev FLUX.1-dev, black-forest-labs/FLUX.1-dev Text image 0.35.2
FLUX.1-Kontext-dev FLUX.1-Kontext-dev, black-forest-labs/FLUX.1-Kontext-dev Text + one image image 0.35.2
FLUX.2-dev FLUX.2-dev, black-forest-labs/FLUX.2-dev Text; optional images image 0.36.0
FLUX.2-klein-4B FLUX.2-klein-4B, black-forest-labs/FLUX.2-klein-4B Text; optional images image 0.37.0
FLUX.2-klein-9B FLUX.2-klein-9B, black-forest-labs/FLUX.2-klein-9B Text; optional images image 0.37.0
HunyuanVideo HunyuanVideo, tencent/HunyuanVideo Text video Not declared
Hunyuanvideo-1.5 Hunyuanvideo-1.5, tencent/HunyuanVideo-1.5, hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-480p_i2v, hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-480p_t2v, hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-720p_i2v, hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-720p_t2v t2v: text; i2v: text + one image (checkpoint must match task) video 0.36.0
Hunyuanvideo-1.5-Distilled Hunyuanvideo-1.5-Distilled, tencent/HunyuanVideo-1.5-Diffusers-720p_i2v_distilled, hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-720p_i2v_distilled Text + one image (i2v) video 0.36.0
Hunyuanvideo-1.5-Sparse Hunyuanvideo-1.5-Sparse, tencent/HunyuanVideo-1.5-Sparse, tencent/HunyuanVideo-1.5-Diffusers-720p_i2v_distilled_sparse Text + one image (i2v) video 0.36.0
Ideogram-4 Ideogram-4, ideogram-ai/ideogram-v4, ideogram-ai/ideogram-4-fp8, CalamitousFelicitousness/Ideogram-4-bf16-Diffusers Text image 0.39.0
Krea-2-Raw Krea-2-Raw, krea/Krea-2-Raw, krea/krea-2-raw Text image 0.39.0
Krea-2-Turbo Krea-2-Turbo, krea/Krea-2-Turbo, krea/krea-2-turbo Text image 0.39.0
LingBot-Video-Dense LingBot-Video-Dense, robbyant/lingbot-video-dense-1.3b Text; optional image video Not declared
LingBot-Video-MoE LingBot-Video-MoE, robbyant/lingbot-video-moe-30b-a3b Text; optional image video Not declared
LTX-2 LTX-2, Lightricks/LTX-2 Text video + audio 0.37.0
LTX-2.3 LTX-2.3, dg845/LTX-2.3-Diffusers Text video + audio 0.37.0
LTX-2.5 LTX-2.5, LTX-2.5-distilled, Lightricks/LTX-2.5-Diffusers t2v: text; i2v: text + one image video + audio 0.40.0
LTX-2.5-full LTX-2.5-full t2v: text; i2v: text + one image video + audio 0.40.0
LTX-Video-0.9.7-dev LTX-Video-0.9.7-dev, Lightricks/LTX-Video-0.9.7-dev t2v: text; i2v: text + one image video 0.34.0
Lumina2 Lumina2, Lumina-Image-2.0, Alpha-VLLM/Lumina-Image-2.0 Text image Not declared
MiniMax-H3 MiniMax-H3, MiniMaxAI/MiniMax-H3 Text; t2va: no images; i2va: first frame; l2va: last frame; fl2va: both video + audio 0.40.0
MiniMax-H3-Image MiniMax-H3-Image, MiniMaxAI/MiniMax-H3-Image Text; optional reference images (editing) image source
MiniMax-H3-Ref2VA MiniMax-H3-Ref2VA Text + one or more reference images video + audio 0.40.0
MiniMax-H3-Turbo MiniMax-H3-Turbo Text; t2va: no images; i2va: first frame; l2va: last frame; fl2va: both video + audio 0.40.0
MOSS-Image MOSS-Image, MOSS-Image-Base, OpenMOSS-Team/MOSS-Image-Base Text image 0.38.0
Prism Prism, Prism-preview-alpha, FrancisRing/Prism Text + one image; optional audio description (text) video + audio 0.33.0
Qwen-Image Qwen-Image, Qwen-Image-2512, Qwen/Qwen-Image, Qwen/Qwen-Image-2512 Text image 0.37.0
Qwen-Image-2.1 Qwen-Image-2.1, Qwen/Qwen-Image-2.1 Text; optional images image 0.41.0
Qwen-Image-2.1-Turbo Qwen-Image-2.1-Turbo, Qwen/Qwen-Image-2.1-Turbo Text; optional images image 0.41.0
Qwen-Image-Edit Qwen-Image-Edit, Qwen-Image-Edit-2509, Qwen-Image-Edit-2511, Qwen/Qwen-Image-Edit, Qwen/Qwen-Image-Edit-2509, Qwen/Qwen-Image-Edit-2511 Text + one image; 2509/2511 require one or more images image 0.37.0
SD3.5 SD3.5, stable-diffusion-3.5-large, stabilityai/stable-diffusion-3.5-large Text image Not declared
SkyReels-V2-I2V-14B SkyReels-V2-I2V-14B, SkyReels-V2-I2V-1.3B, Skywork/SkyReels-V2-I2V-14B-540P-Diffusers, Skywork/SkyReels-V2-I2V-1.3B-540P-Diffusers Text + one image video 0.36.0
SkyReels-V2-T2V-14B SkyReels-V2-T2V-14B, Skywork/SkyReels-V2-T2V-14B-540P-Diffusers Text video 0.36.0
VDN-H3 VDN-H3, OpenVDN/vdn-minimax-h3 Text (t2va) video + audio 0.40.0
Wan2.1-I2V Wan2.1-I2V, Wan-AI/Wan2.1-I2V-14B-720P-Diffusers Text + one image video 0.35.2
Wan2.1-T2V Wan2.1-T2V, Wan-AI/Wan2.1-T2V-14B-Diffusers Text video 0.35.2
Wan2.1-T2V-1.3B Wan2.1-T2V-1.3B, Wan-AI/Wan2.1-T2V-1.3B-Diffusers Text video 0.35.2
Wan2.1-VACE-14B Wan2.1-VACE-14B, Wan2.1-VACE-1.3B, Wan-AI/Wan2.1-VACE-14B-diffusers, Wan-AI/Wan2.1-VACE-1.3B-diffusers Text + first and last frames (two images) video 0.35.2
Wan2.2-Distilled-I2V Wan2.2-Distilled-I2V Text + one image video 0.35.2
Wan2.2-I2V Wan2.2-I2V, Wan-AI/Wan2.2-I2V-A14B-Diffusers Text + one image video 0.35.2
Wan2.2-T2V Wan2.2-T2V, Wan-AI/Wan2.2-T2V-A14B-Diffusers Text video 0.35.2
Wan2.2-TI2V Wan2.2-TI2V, Wan-AI/Wan2.2-TI2V-5B-Diffusers t2v: text; i2v: text + one image video 0.35.2
Z-Image Z-Image, Tongyi-MAI/Z-Image Text image 0.36.0
Z-Image-Turbo Z-Image-Turbo, Tongyi-MAI/Z-Image-Turbo Text image 0.36.0

Some models outside this registry have legacy example scripts. These scripts use the legacy parser; their support is not included in these runner tables.