perf(zerogpu): cap generate_image GPU slot at duration=45 for better scheduling

fix(safety): add midget to MINOR_NEG steer

feat: add support/follow/discord footer line

fix(progress): stream per-step progress for anima image models

fix(prompt): strip web prompt-helper @ markers before diffusion

fix(safety): minor flag wording -> child markers + adult-exclusion (Variant B)

feat(safety): standalone moderate_image endpoint + multi-subject minor flag

feat(safety): adultify realistic prompts, strengthen minor-block negatives, dedicated Qwen guard VLM

feat(models): load wai-anima/samanima as Anime/Realistic, unify prefix

feat: restructure R2 key to namespace/uid/timestamp-rand format

chore: remove temporary moderate_debug tuning endpoint

tmp: add moderate_debug endpoint for rubric tuning (to be removed)

feat: stricter gemma4 moderation rubric to stop under-rating explicit content

feat: watermark assets + skip R2 upload for web-UI generations

Anima loader: fix cosmetic tensor-count log + assert DiT fully materialized

Fix Anima loader: de-prefix CivitAI checkpoint so finetuned DiT+adapter actually load

Wire pornmaster-anima (Anima) into the image-model registry with prefix + negative

Per-model prefix + default_negative (realism-illustrious, wai-anima, samanima); apply in illustrious+anima branches

Image-model registry: IMAGE_MODELS env + illustrious/zimageturbo/anima families

Fix moderation under-rating: rubric + flag-derived rating floor

Prompt-improve as system prompt + One Obsession quality defaults

Switch assistant back to gemma-4-E4B (bf16)

Switch assistant back to Qwen3.5-9B-Uncensored (bf16)

Revert assistant to gemma-4-E4B (bf16) — 12B 8-bit was too slow

Switch assistant to gemma-4-12B (8-bit) + expose system prompt

swap assistant VLM -> ccharnkij/Qwen3.5-9B-Uncensored (test explicit-rating)

feat: switch assistant VLM Qwen3.5-4B -> gemma-4-E4B-it-Uncensored-MAX

fix: VLM assistant never stops (no generation_config) -> pin eos/pad tokens

feat: expose negative prompt in UI, gate it on model capability

fix: stop VLM assistant looping into long repeated output

feat: optional prompt_improve_instruction for /generate_image

feat: image_to_video_assets endpoint (image -> motion prompt + image-context check)

feat: per-image moderation check + WebP upload pipeline

feat: add UI-less prompt_to_video_assets endpoint (LLM->image->LLM)

feat: stream generation + assistant progress over SSE

feat: add Reasoning On/Off toggle to Prompt Assistant (Qwen enable_thinking)

fix: drop unsupported show_copy_button kwarg on Textbox (Gradio 6)

fix: add torchvision for Qwen3.5-VL video processor backend

feat: add Qwen3.5-4B Prompt Assistant tab (text + image, single-turn)

feat: log total GPU time consumed per generate_image call

fix: pad SDXL compel conditionings for long prompts

feat: switch both pipelines to custom dataset checkpoints

fix: drop unsupported show_api kwarg on prompt.submit (Gradio 6)

feat: upload generated images to R2 with prompt/params + requester uid in metadata

Fix RUNTIME_ERROR (kernels/transformers import) and add NoobAI-XL model selector

Set output image format to PNG for higher quality (#56)

fixed

debug

allow MCP (#39)

Update app.py