Improve default example prompts

Reframe StepX as guided product workflow

Overlay DAAM maps on denoising step images

Add guided explanations to StepX UI

Fix model settings left alignment by wrapping dropdown in Row

Redesign model settings layout: dropdown full-width, button+status below

Fix Load Model button height to fill container and align with dropdown

Fix model row alignment: remove dropdown label, add HTML label above row

Fix Load Model button alignment with flexbox justify-content: flex-end

Fix model dropdown and Load Model button alignment using label spacer

Fix model settings UI alignment and rename button to Load Model

Fix model settings UI: widen dropdown, move status below row

Fix TITAN annotation labels always showing 'object' (ZeroGPU fork isolation)

Fix ZoeDepth nested hub load trust prompt (intel-isl/MiDaS)

Fix tokenizer access and ZoeDepth trust_repo for ZeroGPU

Fix PicklingError: make step_data and GlobalHeatMap picklable for ZeroGPU

Return state from @spaces.GPU fns; write to app_state in parent process

Clarify Model Settings UX: collapsed by default, renamed button

Fix all nested @spaces.GPU load calls — use _load_model_impl() instead

Fix device always being cpu: re-detect inside each @spaces.GPU function

Fix shared worker state bugs + pre-download stanza at startup

Fix TITAN annotation crash: return [] not None when no contours found

Fix card background for dark mode — use rgba + CSS vars

Bundle TITAN source; fix Jupyter-only imports; add pycocotools

Redesign UI: new title, header cards, violet theme, model settings open

Add stanza and TITAN package to enable TITAN tab

Remove audioop-lts — ZeroGPU uses Python 3.10 which has audioop built in

Upgrade Gradio to >=5.25.0 to fix API schema TypeError

fix: upgrade to Gradio 5, remove deprecated queue= params, fix huggingface_hub compat

fix: use audioop-lts for Python 3.13 audioop compatibility

fix: add pyaudioop for Python 3.13 compatibility (audioop removed)

fix: bundle daam locally, remove all transformers version conflicts

fix: bundle daami2i locally, unpin transformers/tokenizers for Python 3.13 compat

fix: pin tokenizers==0.15.2 to avoid Rust build requirement

Initial deployment: StepX attention visualization