Prompt Builder: add pose reference image upload -> auto ControlNet OpenPose lock (reuses pose path), prompt drives face/clothes/scene (#38)

Prompt Builder: ท่าโพส -> multiselect dropdown of SD1.5-reliable pose tags (stand/sit/lie/action/hands) (#37)

Prompt Builder: มุมกล้อง/ช็อต -> multiselect dropdown of reliable shot/angle tags (full body/wide trigger canvas bump) (#36)

Prompt Builder: รูปร่าง/สัดส่วน -> multiselect dropdown of model-reliable body tags (Thai label, English value) + free text; _field() handles list values (#35)

Framing+gaze fix: detect full-body from original Thai (translator drops it) + inject English tag, canvas 896; swap looking-at-camera->looking at viewer (#34)

Fix scene truncation: front-load scene/lighting/composition in build_prompt + tighter compact-tag Typhoon output (fewer tokens, keep location) (#33)

Full-body fix: shot field first, shorter style_prefix (free CLIP tokens), auto-tall canvas 512x832 on full-body intent (sd15) (#32)

Prompt Builder: add สีผิว (skin tone) + โครงหน้า (face shape) fields (#31)

majicMIX photoreal pass: drop art tokens (masterpiece/best quality), add photographic style_prefix + sd-vae-ft-mse VAE, photo-focused Typhoon prompt, CFG 5.0 (#30)

majicMIX anatomy fix: hires denoise 0.4->0.3 + front-loaded anatomy negatives (extra legs/limbs/hands) (#29)

Lighten majicMIX hires (scale 1.3, steps 12) to reduce GPU/quota cost (#28)

majicMIX realism boost: hires fix + DPM++ 2M Karras + EasyNegative + lower CFG/detailed prompts (#27)

Add Prompt Builder: fill-in-the-blank form that assembles a prompt (#26)

Add photoreal SDXL models: RealVisXL V5 + Juggernaut XL v9 (fix plastic skin) (#25)

Disable AsianRealistic V6 (all HF mirrors broken); majicMIX default (#24)

AsianRealistic: try noussaler single_file mirror (stablediffusionapi mirror was broken) (#23)

AsianRealistic: load in fp32 to fix NaN rainbow-noise output (#22)

TEMP: AsianRealistic first for diagnosis (#21)

Fix AsianRealistic rainbow noise: override VAE with sd-vae-ft-mse (#20)

Restore model order (majicMIX default) (#19)

Robust LoRA load: fall back to UNet-only keys when kohya TE conversion fails (#18)

TEMP: LoRA first for diagnosis (#17)

Re-enable Asian Girls Face + AsianRealistic V6 via HF mirrors (no Civitai token needed) (#16)

Remove clip_skip; hide token-gated models (Civitai+FLUX); clearer Civitai error (#15)

Typhoon as default translator + Pose (ControlNet) for SDXL (#14)

Add Pose lock mode (ControlNet OpenPose) for SD1.5 (#13)

Fix Typhoon translation (standard generate pattern + better error logs) (#12)

Add Thai->English auto-translate (NLLB-200 + Typhoon 2 selectable) (#11)

Add majicMIX preview thumbnail (#10)

Make clip_skip best-effort (retry without it on version incompat) (#9)

Quality fix: majicMIX model, Euler a, CLIP skip 2, 512x768 default, stronger negatives (#8)

Real model previews: generated SD1.5/SDXL samples + official FLUX grid (#7)

Add model preview images + image gallery picker (#6)

Delete __pycache__/app.cpython-310.pyc

Revert to modern Gradio (5.9.1) + keep bool-schema monkeypatch (#5)

Monkeypatch gradio_client to fix api_info bool-schema crash (#4)

Pin coherent gradio-4.44.1-era deps (hub 0.25.2, diffusers 0.31) (#3)

Pin Gradio to 4.44.1 to fix API schema bug (#2)

Add Character Studio app, registry, requirements, docs (#1)

initial commit