Hugging Face text-to-image shortlist ad68d33de7200cf3: adapters, quantized packages and text-encoder metadata

This discovery guide covers all 150 entries in the retained three-page Hugging Face text-to-image inventory ad68d33de7200cf3+2, listed on 2026-10-28 at 01:33:26-27 UTC. It supports a shortlist for further inspection. The archived evidence is a derived metadata inventory: model README bodies, weight files and actual conditioning interfaces were not inspected, and the original API response bodies are not members of this archive. The listing is download-sorted; its order and counts are time-specific observations, not quality rankings or a complete Hub census. The task label does not establish that each entry is a standalone generator. The following exact metadata relationships give different follow-up routes: | Retained repository | Recorded relationship or field | What still needs inspection | |---|---|---| | diffusers/controlnet-depth-sdxl-1.0 | controlnet tag; base_model:adapter:stabilityai/stable-diffusion-xl-base-1.0 | Base compatibility, required depth/control inputs and the actual invocation | | akhaliq/MOSS-Image-SFT-4-steps | lora and adapter tag for OpenMOSS-Team/MOSS-Image-SFT; base_model also lists lodestones/MOSS-Image-SFT-8-steps-dmd | Full dependency/merge recipe; the name does not establish a validated four-step run | | QuantStack/Qwen-Image-GGUF | quantized relationship to Qwen/Qwen-Image; library_name gguf | Loader, exact weight files and quantization behavior; no VRAM or quality result is supplied | | TechnoBaptist/Qwen-Image-2.1-Turbo-Uncensored-GGUF | text-encoder tag; base_model Qwen/Qwen3-VL-8B-Instruct | Its component role and integration; the Qwen-Image name alone does not identify a complete image model | | nunchaku-ai/nunchaku-qwen-image-edit-2509 | image-editing and SVDQuant tags; quantized relationship to Qwen/Qwen-Image-Edit-2509 | Actual image inputs, supported implementation and component revisions | | Phil2Sat/Qwen-Image-Edit-Rapid-AIO-GGUF | library_name comfyUI; quantized relationship to Phr00t/Qwen-Image-Edit-Rapid-AIO; base_model additionally lists Qwen/Qwen-Image-Edit-2509 and lightx2v/Qwen-Image-Lightning | What is bundled versus separately required, exact controls and compatible workflow | Treat base_model lists and generated relation tags as separate recorded fields. For example, the MOSS and Phil2Sat rows list multiple bases but expose a narrower adapter or quantized relation tag. That does not establish an executed dependency graph or a merge recipe. Missing base_model metadata likewise does not prove an original standalone model. Editing discovery crosses the text-to-image filter: Qwen/Qwen-Image-2.1-Turbo records image-editing and diffusers:QwenImage21Pipeline tags; chriswritescode/Turbo8-LoRA-Qwen-Image-2.1 records image-editing plus an adapter relation; mlx-community/Qwen-Image-2.1-mflux-q4 records an image-to-image tag and library_name mflux. These are leads to inspect, not evidence that the same prompt, image count, masks, reference roles or controls work across them. The inventory contains no submitted input schema, edit-faithfulness comparison or measured runtime. Keep the exact listed repository SHA with each lead. Representative metadata pins are: - diffusers/controlnet-depth-sdxl-1.0: 17bb97973f29801224cd66f192c5ffacf82648b4. - akhaliq/MOSS-Image-SFT-4-steps: 2ded95c37a029360245926fe72e475e30039bbfb. - QuantStack/Qwen-Image-GGUF: 257f261fa92593bed760aa6fa3f7921a49fea00f. - TechnoBaptist/Qwen-Image-2.1-Turbo-Uncensored-GGUF: f184390dce1f6b6c9bb9cb288945c9619c10cef4. - nunchaku-ai/nunchaku-qwen-image-edit-2509: e93a5fb77403d02a5a73c7cc8707b292c6ebc659. - Phil2Sat/Qwen-Image-Edit-Rapid-AIO-GGUF: 02a3fc559cbb2231a052bd743138db86fa4f2dcb. Every entry supplies a 40-character repository revision, but its card bytes/hash remain unknown. A listed revision is a future capture address, not proof that its README or weights were read. Preserve the full inventory, including unselected entries; this table illustrates different investigation routes rather than endorsing six products. Access and terms need separate checks. In this inventory, 147 entries report gated=false and three report gated=auto. Sixteen licence values are null, 49 are other and three are unknown. An ungated observation is not a licence grant. Qwen/Qwen-Image-2.1-Turbo and chriswritescode/Turbo8-LoRA-Qwen-Image-2.1 both declare other, license_name qwen-research and license_link LICENSE; those are pointers to unread terms. The mflux-q4 row declares apache-2.0 while its recorded base is Qwen/Qwen-Image-2.1: the inventory alone cannot establish how those declarations relate or whether they cover all components. Next inspect the chosen repository's pinned card and licence, its declared bases/adapters and actual conditioning API or workflow. Keep card declarations distinct from code, weights, encoders, VAE, services, input media and output terms. No permission transfers through a base_model link, and no existing terms hold is cleared here. Metadata inspection only; no model run, output review, measured quality, memory requirement or independent adoption.

{
  "citations": [
    {
      "url": "https://huggingface.co/api/models?pipeline_tag=text-to-image&sort=downloads&direction=-1&limit=50&expand=sha&expand=gated&expand=cardData&expand=lastModified&expand=createdAt&expand=downloads&expand=likes&expand=tags&expand=library_name&expand=pipeline_tag&cursor=eyIkb3IiOlt7ImRvd25sb2FkcyI6MzkwNTYsIl9pZCI6eyIkZ3QiOiI2NjdjMzA1YzY0YWFmZGViNDc2N2IzOTEifX0seyJkb3dubG9hZHMiOnsiJGx0IjozOTA1Nn19LHsiZG93bmxvYWRzIjpudWxsfV0sInNlYXJjaFNlcXVlbmNlVG9rZW4iOiJDTERzelFJYUNTRUFBQUFBQUJMalFCb09XZ3htZkRCY1pLcjk2MGRuczVFaURsb01abnd3WEdTcS9ldEhaN09SIn0%3D",
      "note": "Origin listing URL recorded in retained derived inventory; model README bodies and original API responses not inspected by curator."
    }
  ]
}