Skip to content

Model catalog — imagehub 0.1

53 architecture families, 123 published checkpoints. A model is a separately published set of weights; fp16 copies, non-EMA weights and single-file .ckpt releases do not count as separate models.

pipeline("t2i", backend=model_id, model="local/folder") selects the family by catalog id; the model folder determines which weights are used.

Validation levels: full — tensors loaded strictly and parity with the recorded reference proven on CPU in fp32. header_only — tensor names and shapes matched, generation not tested. experimental — not tested with the original weights. Models are full unless noted otherwise.

Weight license governs the separately downloaded checkpoint. Code license and Code commercial use govern this repository's implementation for that row; they default to Apache-2.0 and yes. Gated means the publisher requires access approval before download.

Model id Family Source Revision Origin Validation Gated Weight license Code license Code commercial use Awesome catalog
ldm-text2im-large-256 ldm CompVis/ldm-text2im-large-256 30de525 author full no Apache-2.0 Apache-2.0 yes ldm
sd-v1-1 stable-diffusion CompVis/stable-diffusion-v1-1 574555a author full no CreativeML-OpenRAIL-M Apache-2.0 yes stable-diffusion-1
sd-v1-2 stable-diffusion CompVis/stable-diffusion-v1-2 f519eeb author full no CreativeML-OpenRAIL-M Apache-2.0 yes stable-diffusion-1
sd-v1-3 stable-diffusion CompVis/stable-diffusion-v1-3 159f26f author full no CreativeML-OpenRAIL-M Apache-2.0 yes stable-diffusion-1
sd-v1-4 stable-diffusion CompVis/stable-diffusion-v1-4 133a221 author full no CreativeML-OpenRAIL-M Apache-2.0 yes stable-diffusion-1
sd-v1-5 stable-diffusion stable-diffusion-v1-5/stable-diffusion-v1-5 451f4fe community mirror full no CreativeML-OpenRAIL-M Apache-2.0 yes stable-diffusion-1
dreamshaper-8 stable-diffusion Lykon/dreamshaper-8 a7e52b9 author full no CreativeML-OpenRAIL-M Apache-2.0 yes stable-diffusion-1
segmind-small-sd stable-diffusion segmind/small-sd f8b087e author full no CreativeML-OpenRAIL-M Apache-2.0 yes bk-sdm
segmind-tiny-sd stable-diffusion segmind/tiny-sd cad0bd7 author full no CreativeML-OpenRAIL-M Apache-2.0 yes bk-sdm
bk-sdm-base stable-diffusion nota-ai/bk-sdm-base 0c03b7a author full no CreativeML-OpenRAIL-M Apache-2.0 yes bk-sdm
bk-sdm-small stable-diffusion nota-ai/bk-sdm-small 572238d author full no CreativeML-OpenRAIL-M Apache-2.0 yes bk-sdm
bk-sdm-tiny stable-diffusion nota-ai/bk-sdm-tiny 0364108 author full no CreativeML-OpenRAIL-M Apache-2.0 yes bk-sdm
bk-sdm-v2-base stable-diffusion nota-ai/bk-sdm-v2-base 21915be author full no CreativeML-OpenRAIL-M Apache-2.0 yes bk-sdm
bk-sdm-v2-small stable-diffusion nota-ai/bk-sdm-v2-small ad82c51 author full no CreativeML-OpenRAIL-M Apache-2.0 yes bk-sdm
bk-sdm-v2-tiny stable-diffusion nota-ai/bk-sdm-v2-tiny 68277af author full no CreativeML-OpenRAIL-M Apache-2.0 yes bk-sdm
sd-v2-1 stable-diffusion sd2-community/stable-diffusion-2-1 bb21548 community mirror full no CreativeML-OpenRAIL++-M Apache-2.0 yes stable-diffusion-2
sd-turbo stable-diffusion stabilityai/sd-turbo b261bac author full no sai-nc-community Apache-2.0 yes sdxl-turbo (SD Turbo)
sdxs-512-0.9 stable-diffusion IDKiro/sdxs-512-0.9 c332f05 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes sdxs
lcm-dreamshaper-v7 stable-diffusion SimianLuo/LCM_Dreamshaper_v7 a85df6a author full no MIT Apache-2.0 yes lcm
dreamlite-base dreamlite carlofkl/DreamLite-base a9a0f15 author full no CC-BY-NC-4.0 Apache-2.0 yes dreamlite
dreamlite-mobile dreamlite carlofkl/DreamLite-mobile a48c656 author full no CC-BY-NC-4.0 Apache-2.0 yes dreamlite
sdxl-base-1.0 stable-diffusion-xl stabilityai/stable-diffusion-xl-base-1.0 4621659 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes sdxl
juggernaut-xl-v9 stable-diffusion-xl RunDiffusion/Juggernaut-XL-v9 cf41923 author full no CreativeML-OpenRAIL-M Apache-2.0 yes sdxl
animagine-xl-4.0 stable-diffusion-xl cagliostrolab/animagine-xl-4.0 2b7c1b3 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes sdxl
dreamshaper-xl-1.0 stable-diffusion-xl Lykon/dreamshaper-xl-1-0 41e6644 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes sdxl
sdxl-refiner-1.0 stable-diffusion-xl stabilityai/stable-diffusion-xl-refiner-1.0 5d4cfe8 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes sdxl
sdxl-lightning-2step stable-diffusion-xl ByteDance/SDXL-Lightning c9a24f4 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes sdxl-lightning
sdxl-lightning-4step stable-diffusion-xl ByteDance/SDXL-Lightning c9a24f4 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes sdxl-lightning
sdxl-lightning-8step stable-diffusion-xl ByteDance/SDXL-Lightning c9a24f4 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes sdxl-lightning
hyper-sdxl-1step stable-diffusion-xl ByteDance/Hyper-SD bc08d97 author full no Unspecified Apache-2.0 yes hyper-sd
dmd2-sdxl-4step stable-diffusion-xl tianweiy/DMD2 be22767 author full no CC-BY-NC-4.0 Apache-2.0 yes dmd2
dmd2-sdxl-1step stable-diffusion-xl tianweiy/DMD2 be22767 author full no CC-BY-NC-4.0 Apache-2.0 yes dmd2
sdxl-turbo stable-diffusion-xl stabilityai/sdxl-turbo 7115331 author full no sai-nc-community Apache-2.0 yes sdxl-turbo
ssd-1b stable-diffusion-xl segmind/SSD-1B 60987f3 author full no Apache-2.0 Apache-2.0 yes ssd-1b
segmind-vega stable-diffusion-xl segmind/Segmind-Vega 7714c43 author full no Apache-2.0 Apache-2.0 yes segmind-vega
koala-700m stable-diffusion-xl etri-vilab/koala-700m 5e73a31 author full no Unspecified Apache-2.0 yes koala
koala-1b stable-diffusion-xl etri-vilab/koala-1b 8ef26f3 author full no Unspecified Apache-2.0 yes koala
koala-lightning-700m stable-diffusion-xl etri-vilab/koala-lightning-700m 358e0e3 author full no Unspecified Apache-2.0 yes koala
koala-lightning-1b stable-diffusion-xl etri-vilab/koala-lightning-1b b4114f7 author full no Unspecified Apache-2.0 yes koala
playground-v2 stable-diffusion-xl playgroundai/playground-v2-1024px-aesthetic 737a24e author full no playground-v2-community Apache-2.0 yes playground-v2
playground-v2.5 stable-diffusion-xl playgroundai/playground-v2.5-1024px-aesthetic 1e032f1 author full no playground-v2dot5-community Apache-2.0 yes playground-v2
sd3-medium stable-diffusion-3 stabilityai/stable-diffusion-3-medium-diffusers ea42f8c author full yes Stability AI Non-Commercial Research Community License Apache-2.0 yes stable-diffusion-3
sd3.5-medium stable-diffusion-3 stabilityai/stable-diffusion-3.5-medium b940f67 author full yes Stability AI Community License Apache-2.0 yes stable-diffusion-3
sd3.5-large stable-diffusion-3 stabilityai/stable-diffusion-3.5-large ceddf0a author full yes Stability AI Community License Apache-2.0 yes stable-diffusion-3
sd3.5-large-turbo stable-diffusion-3 stabilityai/stable-diffusion-3.5-large-turbo ec07796 author full yes Stability AI Community License Apache-2.0 yes stable-diffusion-3
flux1-schnell flux black-forest-labs/FLUX.1-schnell 741f7c3 author full yes Apache-2.0 Apache-2.0 yes flux-1
flux1-dev flux black-forest-labs/FLUX.1-dev 3de623f author full yes FLUX.1-dev Non-Commercial License Apache-2.0 yes flux-1
flux1-krea-dev flux black-forest-labs/FLUX.1-Krea-dev 8162a9c author full yes FLUX.1-dev Non-Commercial License Apache-2.0 yes flux-1
flux-1-lite-8b flux Freepik/flux.1-lite-8B 16b9373 author full no FLUX.1-dev Non-Commercial License Apache-2.0 yes flux-1
flux1-kontext-dev flux black-forest-labs/FLUX.1-Kontext-dev 24e9ded author full yes FLUX.1-dev Non-Commercial License Apache-2.0 yes flux-1-kontext
visualcloze-384 visualcloze VisualCloze/VisualClozePipeline-384 59c469d author full no Apache-2.0 Apache-2.0 yes visualcloze
joyai-image-edit joyai-image jdopensource/JoyAI-Image-Edit-Diffusers 4b41fb2 author full no Apache-2.0 Apache-2.0 yes joyai-image
joyai-image-edit-plus joyai-image jdopensource/JoyAI-Image-Edit-Plus-Diffusers c268646 author full no Apache-2.0 Apache-2.0 yes joyai-image
flux2-klein-4b flux2 black-forest-labs/FLUX.2-klein-4B e7b7dc2 author full no Apache-2.0 Apache-2.0 yes flux-2
flux2-klein-9b flux2 black-forest-labs/FLUX.2-klein-9B 92196c8 author full yes FLUX Non-Commercial License Apache-2.0 yes flux-2
hunyuan-dit-v1.1-distilled hunyuan-dit Tencent-Hunyuan/HunyuanDiT-v1.1-Diffusers-Distilled 527cf2e author full no Tencent Hunyuan Community License Apache-2.0 yes hunyuan-dit
hunyuan-dit-v1.2 hunyuan-dit Tencent-Hunyuan/HunyuanDiT-v1.2-Diffusers 5e96094 author full no Tencent Hunyuan Community License Apache-2.0 yes hunyuan-dit
hunyuan-dit-v1.2-distilled hunyuan-dit Tencent-Hunyuan/HunyuanDiT-v1.2-Diffusers-Distilled ba991d1 author full no Tencent Hunyuan Community License Apache-2.0 yes hunyuan-dit
deepfloyd-if-i-xl deepfloyd-if DeepFloyd/IF-I-XL-v1.0 c03d510 author full yes DeepFloyd IF License (non-commercial) Apache-2.0 yes deepfloyd-if
kandinsky-2-1 kandinsky21 kandinsky-community/kandinsky-2-1 04b3666 author full no Apache-2.0 Apache-2.0 yes kandinsky-2
kandinsky-2.2 kandinsky22 kandinsky-community/kandinsky-2-2-decoder 9ae140d author full no Apache-2.0 Apache-2.0 yes kandinsky-2
kandinsky-3 kandinsky3 kandinsky-community/kandinsky-3 bf79e6c author full no Apache-2.0 Apache-2.0 yes kandinsky-3
kandinsky-5-t2i-lite kandinsky5 kandinskylab/Kandinsky-5.0-T2I-Lite-sft-Diffusers 25da1e8 author full no MIT Apache-2.0 yes kandinsky-5
pixart-sigma-xl-1024 pixart PixArt-alpha/PixArt-Sigma-XL-2-1024-MS e102b35 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes pixart-sigma
pixart-alpha-xl-1024 pixart PixArt-alpha/PixArt-XL-2-1024-MS b89adad author full no CreativeML-OpenRAIL++-M Apache-2.0 yes pixart-alpha
pixart-alpha-xl-512 pixart PixArt-alpha/PixArt-XL-2-512x512 50f7021 author full no CreativeML-OpenRAIL++-M Apache-2.0 yes pixart-alpha
pixart-sigma-xl-512 pixart PixArt-alpha/PixArt-Sigma-XL-2-512-MS 76fb7eb author full no CreativeML-OpenRAIL++-M Apache-2.0 yes pixart-sigma
pixart-sigma-xl-256 pixart PixArt-alpha/PixArt-Sigma-XL-2-256x256 495cb1b author full no OpenRAIL (exact terms unspecified) Apache-2.0 yes pixart-sigma
pixart-sigma-xl-2k pixart PixArt-alpha/PixArt-Sigma-XL-2-2K-MS cd274f4 author full no OpenRAIL (exact terms unspecified) Apache-2.0 yes pixart-sigma
pixart-lcm-xl-1024 pixart PixArt-alpha/PixArt-LCM-XL-2-1024-MS e153060 author header_only no CreativeML-OpenRAIL++-M Apache-2.0 yes pixart-delta
sana-1600m-1024 sana Efficient-Large-Model/Sana_1600M_1024px_diffusers ac0da2f author full no Apache-2.0 Apache-2.0 yes sana
sana-600m-1024 sana Efficient-Large-Model/Sana_600M_1024px_diffusers 28f3af7 author full no Apache-2.0 Apache-2.0 yes sana
sana-sprint-1.6b-1024 sana Efficient-Large-Model/Sana_Sprint_1.6B_1024px_diffusers 19683c5 author full no Apache-2.0 Apache-2.0 yes sana
sana-1.5-4.8b-1024 sana Efficient-Large-Model/SANA1.5_4.8B_1024px_diffusers 9468102 author full no Apache-2.0 Apache-2.0 yes sana
kolors kolors Kwai-Kolors/Kolors-diffusers 7e091c7 author full no Apache-2.0 Apache-2.0 yes kolors
lumina-image-2.0 lumina2 Alpha-VLLM/Lumina-Image-2.0 53504ab author full no Apache-2.0 Apache-2.0 yes lumina-image-2
lumina-next lumina-next Alpha-VLLM/Lumina-Next-SFT-diffusers 0ee5ec9 author full no Apache-2.0 Apache-2.0 yes lumina-next
lumina-dimoo lumina-dimoo Alpha-VLLM/Lumina-DiMOO f014549 author full no Apache-2.0 Apache-2.0 yes lumina-dimoo
omnigen-v1 omnigen Shitao/OmniGen-v1-diffusers 016e2f6 author full no MIT Apache-2.0 yes omnigen
chroma1-hd chroma lodestones/Chroma1-HD 0e0c60e author full no Apache-2.0 Apache-2.0 yes chroma1-hd
chroma1-flash chroma lodestones/Chroma1-Flash 093c4f6 author full no Apache-2.0 Apache-2.0 yes chroma1-flash
chroma1-base chroma lodestones/Chroma1-Base 50ebe12 author full no Apache-2.0 Apache-2.0 yes chroma1-base
amused-256 amused amused/amused-256 09b6259 author full no OpenRAIL++ Apache-2.0 yes amused
amused-512 amused amused/amused-512 6f3f309 author full no OpenRAIL++ Apache-2.0 yes amused
meissonic meissonic MeissonFlow/Meissonic 7ee5068 author full no Apache-2.0 Apache-2.0 yes meissonic
auraflow-v0.3 auraflow fal/AuraFlow-v0.3 2cd8588 author full no Apache-2.0 Apache-2.0 yes auraflow
prx-1024-t2i-beta prx Photoroom/prx-1024-t2i-beta 0370504 author full no Apache-2.0 Apache-2.0 yes prx
prx-512-t2i-sft prx Photoroom/prx-512-t2i-sft 2996423 author full no Apache-2.0 Apache-2.0 yes prx
llamagen-t2i-xl-256 llamagen peizesun/llamagen_t2i 276f5c5 author full no MIT Apache-2.0 yes llamagen
llamagen-t2i-xl-512 llamagen peizesun/llamagen_t2i 276f5c5 author full no MIT Apache-2.0 yes llamagen
infinity-2b infinity FoundationVision/Infinity e710892 author full no MIT Apache-2.0 yes infinity
hart hart mit-han-lab/hart-0.7b-1024px b1521ff author full no MIT Apache-2.0 yes hart
show-o-512 show-o showlab/show-o-512x512 dbe757d author full no Apache-2.0 Apache-2.0 yes show-o
emu3-gen emu3 BAAI/Emu3-Gen 40b3bb8 author full no Apache-2.0 Apache-2.0 yes emu3
pixeldit pixeldit nvidia/PixelDiT-1300M-1024px 7c63b99 author full no NVIDIA License (NSCLv1; non-commercial research or evaluation) NSCLv1 no pixeldit
krea2-turbo krea2 krea/Krea-2-Turbo 98e0fe1 author full yes Krea 2 Community License Apache-2.0 yes krea-2
ernie-image ernie-image baidu/ERNIE-Image 5346b31 author full no Apache-2.0 Apache-2.0 yes ernie-image
longcat-image longcat-image meituan-longcat/LongCat-Image d2ea50b author full no Apache-2.0 Apache-2.0 yes longcat-image
hunyuanimage-2-1 hunyuanimage hunyuanvideo-community/HunyuanImage-2.1-Diffusers 7e7b7a1 community-conversion full no Tencent Hunyuan Community License Apache-2.0 yes hunyuanimage-2-1
cogview3-plus-3b cogview3 zai-org/CogView3-Plus-3B 5d70e40 author full no Apache-2.0 Apache-2.0 yes cogview3-plus
cogview4-6b cogview4 zai-org/CogView4-6B 63a52b7 author full no Apache-2.0 Apache-2.0 yes cogview4
ideogram4-fp8 ideogram4 ideogram-ai/ideogram-4-fp8 ee79a72 author full yes Ideogram 4 Non-Commercial License Apache-2.0 yes ideogram-4
qwen-image qwen-image Qwen/Qwen-Image 75e0b4b author full no Apache-2.0 Apache-2.0 yes qwen-image
qwen-image-2.1 qwen-image Qwen/Qwen-Image-2.1 d26bb61 author full no Qwen Research License Apache-2.0 yes qwen-image-2
qwen-image-edit qwen-image Qwen/Qwen-Image-Edit ac7f931 author full no Apache-2.0 Apache-2.0 yes qwen-image-edit
qwen-image-edit-2509 qwen-image Qwen/Qwen-Image-Edit-2509 d3968ef author full no Apache-2.0 Apache-2.0 yes qwen-image-edit
ovis-image-7b ovis-image ATH-MaaS/Ovis-Image-7B 41be1c5 author full no Apache-2.0 Apache-2.0 yes ovis-image
z-image z-image Tongyi-MAI/Z-Image 04cc4ab author full no Apache-2.0 Apache-2.0 yes z-image
z-image-turbo z-image Tongyi-MAI/Z-Image-Turbo f332072 author full no Apache-2.0 Apache-2.0 yes z-image
janus-pro-1b janus deepseek-community/Janus-Pro-1B 1655280 community conversion full no DeepSeek Model License Apache-2.0 yes janus-pro
janus-pro-7b janus deepseek-community/Janus-Pro-7B 8f23139 community conversion full no DeepSeek Model License Apache-2.0 yes janus-pro
bagel-7b-mot bagel ByteDance-Seed/BAGEL-7B-MoT 5019f57 author full no Apache-2.0 Apache-2.0 yes bagel
stable-cascade stable-cascade stabilityai/stable-cascade a89f66d author full no stable-cascade-nc-community Apache-2.0 yes wuerstchen
stable-cascade-lite stable-cascade stabilityai/stable-cascade a89f66d author full no stable-cascade-nc-community Apache-2.0 yes wuerstchen
wuerstchen-v2 wuerstchen warp-ai/wuerstchen c3da414 author full no mit Apache-2.0 yes wuerstchen
hidream-i1-fast hidream HiDream-ai/HiDream-I1-Fast 4856a5d author full yes MIT Apache-2.0 yes hidream-i1
hidream-i1-dev hidream HiDream-ai/HiDream-I1-Dev 0fad2ea author full yes MIT Apache-2.0 yes hidream-i1
hidream-i1-full hidream HiDream-ai/HiDream-I1-Full 8ccbbfb author full yes MIT Apache-2.0 yes hidream-i1
hidream-o1-image hidream-o1 HiDream-ai/HiDream-O1-Image 0b0901d author full no MIT Apache-2.0 yes hidream-o1-image
glm-image glm-image zai-org/GLM-Image 2c433cc author full no MIT Apache-2.0 yes glm-image
cosmos-predict2-2b-t2i cosmos2 nvidia/Cosmos-Predict2-2B-Text2Image acdb5fd author full yes NVIDIA Open Model License Apache-2.0 yes cosmos-predict2-2b-t2i
nextstep-1-large nextstep stepfun-ai/NextStep-1-Large 97dd0b9 author full no Apache-2.0 Apache-2.0 yes nextstep-1
nucleus-image nucleus-image NucleusAI/Nucleus-Image 5e963db author full no Apache-2.0 Apache-2.0 yes nucleus-image

The original repository of sd-v1-5 (runwayml/stable-diffusion-v1-5) has been removed; the repository used is a community mirror. The v1.1–v1.3 fp32 weights are published only as .bin files, so they are loaded with weights_only=True, containing tensors only. Segmind Small-SD and Tiny-SD likewise publish tensor-only .bin component files. Small-SD's tensors are float32; Tiny-SD has only float16 published tensors, which are upcast for CPU float32 validation. Juggernaut XL v9 publishes only float16 component files. Animagine XL 4.0 likewise publishes float16 component tensors, although its component filenames do not carry an fp16 suffix. Both are upcast for CPU float32 validation. DreamShaper XL 1.0 publishes float32 component tensors. Cosmos-Predict2 is gated and requires accepting the NVIDIA Open Model License before download. Its transformer and VAE are float32; its two T5 encoder shards are bfloat16 and are upcast for CPU float32 inference.

Composite checkpoints can name a base model, omit base files with base_exclude, and overlay their own files. They can also use subfolders to place files from a second repository in a named snapshot subfolder. DMD2 excludes the base U-Net safetensors because its replacement U-Net is a PyTorch pickle under a different name. Kandinsky 2.1 and 2.2 place their pinned prior repositories in the snapshot's prior/ subfolder while each decoder remains at the snapshot root. The PixArt-Sigma 256, 512 and 2K repositories publish only a transformer, so those models use pixart-sigma-xl-1024 as their base and overlay their transformer files with their own. Stable Cascade does the same with stabilityai/stable-cascade-prior at revision 7ca32c2, and Würstchen v2 with warp-ai/wuerstchen-prior at revision c41be63. python -m imagehub.pretrained downloads both sources, each pinned and verified, and the snapshot manifest records provenance for every file.

LlamaGen uses the same mechanism for its text encoder: google/flan-t5-xl at revision 7d6315d goes into the snapshot's text_encoder/ subfolder, while the GPT and VQ-16 PyTorch checkpoints of peizesun/llamagen_t2i (format pytorch, loaded with weights_only=True) stay at the root. The Flan-T5 shards also hold the decoder, which imagehub never reads.

Infinity-2B also places google/flan-t5-xl at revision 7d6315d in text_encoder/. infinity_2b_reg.pth and infinity_vae_d32reg.pth remain at the snapshot root and are loaded as weights-only PyTorch archives; training-only VAE perceptual tensors are verified but not loaded.

The hyper-sdxl-1step model card has no license metadata. The ByteDance/Hyper-SD repository's license file covers only the FLUX-related files, so the Hyper-SDXL weights are recorded as Unspecified.

The dmd2-sdxl-4step and dmd2-sdxl-1step weights are licensed under CC-BY-NC-4.0. The license prohibits commercial use; see the license before downloading or using the weights.

The dreamlite-base and dreamlite-mobile weights are licensed under CC-BY-NC-4.0. The license prohibits commercial use; see the license before downloading or using the weights.

The catalog uses the deepseek-community/Janus-Pro-1B and deepseek-community/Janus-Pro-7B transformers conversions with community-conversion provenance because the authors' repositories ship weights for the original janus package rather than the transformers layout.

All four Stable Diffusion 3 and 3.5 repositories are gated. Accept their terms on Hugging Face, then set HF_TOKEN when downloading with python -m imagehub.pretrained. SD3 Medium uses the non-commercial research license. The SD3.5 Community License allows registered commercial use below USD 1 million annual revenue and requires an enterprise license at or above that threshold.

The Black Forest Labs FLUX.1 repositories are gated. Accept their terms on Hugging Face, then set HF_TOKEN when downloading with python -m imagehub.pretrained. Flux.1-Lite is ungated. The schnell weights use Apache-2.0; the dev, Krea, Lite and Kontext weights use the FLUX.1-dev Non-Commercial License.

The flux2-klein-9b repository is gated. Accept its terms on Hugging Face, then set HF_TOKEN when downloading with python -m imagehub.pretrained.

The HunyuanDiT weights use the Tencent Hunyuan Community License, which excludes use in the European Union, the United Kingdom and South Korea and requires a separate license above 100 million monthly active users; see the license.

The krea2-turbo repository is gated. Accept its terms on Hugging Face, then set HF_TOKEN when downloading with python -m imagehub.pretrained. The weights use the Krea 2 Community License; commercial use is conditional below USD 1 million annual revenue and requires an enterprise license at or above that threshold.

krea2-turbo is the distilled inference checkpoint. Krea 2 LoRA training uses the separate, gated krea/Krea-2-Raw checkpoint at 6b0ece7fffb640c5e3bcbe0a7f10f66b8e60a603; accept the same license before downloading it. Raw-trained adapters run on Turbo and remain bound by the Krea 2 Community License.

The ideogram4-fp8 repository is gated. Accept its terms on Hugging Face, then set HF_TOKEN when downloading with python -m imagehub.pretrained. The weights use the Ideogram 4 Non-Commercial License; see the license.

Every official Ideogram 4 checkpoint is quantized, so ideogram4-fp8 remains inference-only.

Kandinsky 5 publishes bfloat16 transformer and text-encoder weights; the catalog lists them as released and the native loaders upcast them to float32 for parity testing.

Kolors publishes float16 weights only (*.fp16.safetensors), so its file list uses those names. Parity is measured on their float32 upcast, which is also what diffusers computes with from_pretrained(..., variant="fp16", torch_dtype=torch.float32).

The hidream-i1-fast, hidream-i1-dev and hidream-i1-full entries combine their respective HiDream-I1 repositories (MIT) with meta-llama/Llama-3.1-8B-Instruct at revision 0e9e39f, placed in the snapshot's text_encoder_4/ subfolder together with its tokenizer. The Llama repository is gated: accept its terms (Llama 3.1 Community License) on Hugging Face, then set HF_TOKEN when downloading with python -m imagehub.pretrained. The HiDream repository does not ship that encoder, and the pipeline loads it separately.

The stable-cascade and stable-cascade-lite weights use the Stable Cascade Non-Commercial Community License (stable-cascade-nc-community), which prohibits commercial use; see the license before downloading or using the weights. Both models share one decoder repository and one prior repository and differ only in the stage C and stage B files they select (prior and decoder, or prior_lite and decoder_lite).