Update Space short_description

Rescale before image to after's exact dims for slider alignment

Stop resizing output to match input dims — keep pipeline-native resolution

Drop anon gen allowance to 1 to protect the ZeroGPU anon pool

Bump anon gen allowance from 1 to 2 before login gate

Gate anon users to login after their first gen

Stop passing width/height to the Qwen pipeline — fixes the centered crop

Use a separate HF_WRITE_TOKEN for gallery dataset pushes

Wire up the 'download the model' link to the public LoRA repo

LoRA filename V2 → V1 to match the renamed repo

Bump to v1.0, point at the renamed LoRA repo

Reposition figure callouts to land on articulation points

Move SOFT-GOODS · PLASTIC onto the horse body

Re-map callouts using actual figure-stage geometry (48% wide, right:4%)

Nudge ARM + KNEE dots right by 5%

Reposition callouts onto her arm, knee, and the horse body

Move ARM + KNEE callouts down to land on her actual arm + knee

Drop fig-08 (TRACKSIDE) + fig-09 (RECORDER), shift 10-12 up to 08-10

Resize output to match input dimensions

Fix callout positions, Qwen 32-multiple resize, gallery dataset uploads

feat: session-based result persistence — recover a gen after tab-away

ui: remove upload size readout from processing modal

ui: lasso cowgirl hero + repositioned callouts + download arrow

ui: re-curated marquee — sports/lifestyle figure set

ui: rename marquee header → Example outputs

ui: cowgirl hero + 12 portrait figures, 9:16 cards, systms.ai link

fix: drop margin auto that collapsed flex compare element

ux: replace celebrity figures with new AI-generated set + placeholder slots

ux: toy-factory copy + dynamic aspect-ratio on result compare (no more crop)

fix: use @gradio/client per HF guidance — does ZeroGPU handshake (issues fresh

fix: capture request via middleware contextvar (Server treats fn args as

debug: forward x-ip-token + log auth signals server-side

fix: type-hint Request on /me; attach OAuth routes to gradio.Server

revert to gradio.Server + add HF OAuth login pill for auth'd ZeroGPU quota

fix: restore gr.mount_gradio_app (broken by overzealous rename)

fix: call demo.queue() so HF detects ZeroGPU function on mounted demo

fix: put @spaces.GPU directly on action so HF detects ZeroGPU function

fix: start uvicorn so the FastAPI+Gradio app actually serves

fix: mount Gradio at /gradio for proper

fix: handshake /config with token + include

debug: log whether __sign token was

fix: forward HF __sign JWT as x-ip-token so ZeroGPU identifies

debug: log entry/exit + full traceback

fix: resize images client-side to

fix: bypass @gradio/client

fix: launch gradio.Server at module top

fix: use jsdelivr @gradio/client to match

Bump gradio to 6.10.0+ (required for

Refactor to gradio.Server pattern

Force bind port 7860