Fetching metadata from the HF Docker repository...
Sep 3 23:10 Tighten wide overhead vehicle footprint Require explicitly wide top-down vehicles to be clearly wider than tall. This rejects the latest public oblique two-wheel rover at aspect 0.947 while preserving the existing fail-closed behavior. Sep 1 15:53 Reject front-heavy balanced overhead vehicles Reject front-facing vehicle cutouts when an explicit top-down brief requires paired features to be equally visible. The exact public rover false positive scores 0.305 rotational balance and is rejected by this narrow fail-closed check. Aug 30 22:19 Document radial overhead validation 22:18 Reject asymmetric radial overhead sprites Aug 29 17:25 Reject incompatible overhead silhouettes Aug 27 08:04 Strengthen strict overhead sprite prompting 04:08 Upgrade generation to full-quality SDXL 03:17 Enforce top-down sprite perspective Aug 26 19:09 Improve generated image sharpness 18:32 Start generation form without sample prompts Aug 25 00:14 Extract transparent assets with BiRefNet 00:06 Reject non-studio cutout attempts Aug 24 23:59 Reject scene-contaminated transparent assets Aug 23 20:19 Support fully custom asset manifests Aug 21 03:29 Normalize generated sprite scale 03:14 Reject miniature duplicate sprites 02:51 Detect touching duplicate sprite subjects 02:33 Scale ZeroGPU duration by asset count 02:20 Keep generation prompts inside CLIP budget 01:10 Generate separate roles and reject multi-view sprite sheets 00:36 Switch to blue and yellow light theme 00:22 Redesign Space interface Aug 18 14:28 Enforce single-subject neural assets Strengthen direct text-to-image prompts and negative prompts to reject contact sheets, turnarounds, duplicates, lineups, and alternate views. 14:20 Document direct text-to-image generation Document that the production model receives written prompts only and no procedural image guide. 14:19 Generate assets directly from text prompts Remove procedural image guides from the production diffusion path. Generate sprites and backgrounds directly from written prompts, keep fail-closed neural enforcement, and make realistic style parsing negation-aware. 07:28 Update README.md Document higher-detail sprite settings 07:28 Update app.py Increase neural sprite detail 07:22 Update README.md Document camera-guided neural sprites 07:21 Update app.py Guide neural sprites by camera and palette 07:13 Update README.md Document the free model-first ZeroGPU pipeline 07:13 Update app.py Make SSD-1B ZeroGPU the required image generator 07:13 Update requirements.txt Prepare dependencies for required ZeroGPU diffusion Aug 17 20:26 Fix procedural subject and background matching Route player sprites by requested subject, remove accidental particle-style triggers, keep shooter backgrounds theme-specific, add a desert overhead renderer, and report that the free default is procedural rather than neural. 07:39 Enforce camera perspective in background generation Aug 16 10:11 Make validation aware of image model behavior 10:02 Add deterministic asset integration workflow Jun 17 02:05 Start roles and theme blank Jun 16 02:01 Add safe HF token validity check 00:35 Start with empty paste-friendly HTML input Jun 13 03:03 Clean up forest background artifacts 02:57 Keep backgrounds separate and prevent player trails Jun 12 03:30 Prevent missing images from black-screening canvas games Jun 10 02:43 Use whole 2D sprite generator for sprite roles 02:36 Prevent sprite generations from becoming texture maps 02:01 Force free diffusion model to float32 on CPU 01:56 Add free local diffusion dependencies 01:55 Use free local Diffusers image model 01:32 Add huggingface_hub dependency 01:32 Use HF Inference Providers 01:00 Show Hugging Face model errors 
