gatilin/deepgen_1_0_sd3_dit
Updated
DiT & Visual Generation
Diffusion transformers, conditional decoders, and generation models, with their conditioning, codec, and sampling requirements.
Note Complete DeepGen-1.0 SD3 DiT and six-layer connector. The language backbone and VAE are excluded. Native conditioning and scheduling interfaces are required.
gatilin/uniddt_512_visual_components
Updated • 26Note UniDDT-512 Noisy ViT and separate diffusion decoder. The Noisy ViT consumes FLUX VAE latents rather than RGB. Native Qwen3-VL conditioning and an external VAE are required.
gatilin/unipic2_metaquery_grpo_9b_dit
UpdatedNote UniPic2 Metaquery GRPO SD3 DiT and 24-layer conditioner. The full language model and VAE are excluded; native reference-image editing interfaces are retained.
gatilin/unispace_pr_dinov2_class_conditional_dit
Updated • 33Note UniSpace PR-DINOv2 class-conditional DiT and normalization statistics. This is an ImageNet-class generator, not the complete multimodal MoT or a text-conditioned generator.
gatilin/open-genception-1.3b-pytorch
1B • Updated • 1Note PyTorch conversion of the authors' GenCeption 1.3B multitask checkpoint: Wan DiT, token heads, and official Wan VAE. Intended for geometry/depth tasks; this is not a Specialist-S benchmark checkpoint from the paper.
gatilin/open-genception-14b-pytorch
14B • UpdatedNote PyTorch conversion of the authors' GenCeption 14B multitask checkpoint: Wan DiT, token heads, and Wan VAE. This is not a Specialist-S benchmark checkpoint; native grouped-attention interfaces are required.
gatilin/open-vision-banana-sana_1600m_multitask
Text-to-Image • 2B • UpdatedNote Open-Vision-Banana SANA multitask image-generation package. This is a generation-task model, distinct from standalone ViT or tokenizer exports.
gatilin/internvlu_image_dit
2B • Updated • 23Note Complete InternVL-U 20-layer generation DiT, including native conditioning and padding interfaces. The language backbone, vision tower and VAE are excluded. Preserve DPM flow sampling and three-branch CFG; validation uses synthetic CPU component fixtures.