gatilin/harmon_0_5b_visual_mar_codec

Image-to-Image • 0.3B • Updated 7 days ago

Note Complete Harmon visual MAR, VAE, diffusion prediction/sampling head, and language-space projections. The Qwen language backbone is excluded.

gatilin/hidream_o1_vision_pixel_io

0.6B • Updated 10 days ago • 25

Note HiDream-O1 vision/DeepStack modules, pixel input, and timestep/output interfaces. The shared language/flow backbone is excluded, so this package does not generate images independently.

gatilin/illumex_vision_flow_io

Image Feature Extraction • 0.5B • Updated 9 days ago

Note ILLUME-X SigLIP vision tower, connector, flow interfaces, and its upstream BAGEL VAE dependency. The language/MoT backbone is excluded.

gatilin/internlumina_u2_visual_io

0.6B • Updated 10 days ago • 32

Note InternLumina-U2 multi-codebook input/output interfaces and AR head. The AToken tokenizer and shared diffusion backbone are separate dependencies.

gatilin/mobile_o_0_5b_visual_components

Image Feature Extraction • 1B • Updated 5 days ago • 1

Note FastViT/MobileCLIP, MCP, complete SANA, and DC-AE encoder/decoder. The Qwen language backbone is excluded. This is a mixed visual-component package, not a ViT-only or VAE-only export.

gatilin/ovis_u1_3b_visual_components

Image Feature Extraction • 2B • Updated 6 days ago

Note AIMv2 soft visual tokenizer, visual embedding table, and complete Yak flow/VAE branch. The understanding tokenizer is not a pixel VQ codec. The Qwen3 backbone is excluded.

gatilin/pixelumm_pixel_io

52.5M • Updated 10 days ago • 29 • 1

Note PixelUMM image/video pixel input and output interfaces. The shared MoT backbone is excluded; this is not a standalone VAE or semantic encoder.

gatilin/sensenova_u1_5_8b_pixel_io

80.7M • Updated 10 days ago • 31

Note Native RGB input/output projections and timestep/noise interfaces. The shared MoT backbone is excluded; this is an interface bundle, not a standalone ViT, VAE, or generator.

gatilin/sensenova_vision_perception_io

Image Feature Extraction • 0.5B • Updated 9 days ago

Note SenseNova-Vision SigLIP tower, flow interfaces, and BAGEL VAE, distinct from SenseNova-U1.5. The shared MoT backbone is excluded.

gatilin/showo2_1_5b_visual_flow_components

Feature Extraction • 1B • Updated 7 days ago

Note Show-o2 semantic tower, fusion interfaces, ten-layer flow branch, and native Wan2.1 VAE. Complete generation still requires the matching Qwen backbone.

gatilin/unicom_vision_compression

0.5B • Updated 10 days ago • 27

Note UniCom semantic compression/expansion and flow interfaces. The pixel decoder, FLUX VAE, and language backbone are excluded. The native aligner input-width mismatch is explicitly rejected rather than implicitly adapted.

gatilin/unilip_1b_visual_components

Image Feature Extraction • 1B • Updated 6 days ago

Note InternVision, connector, custom DC-AE decoder, and complete SANA branch. Image encoding uses the vision tower rather than an RGB VAE. Loading requires the separately obtained upstream decoder source specified in the model card.