back to a gradio CPU Space, still just a signpost

open links outside the Spaces iframe

point at the merged Space, drop the runtime that moved with it

Switch to free color

Remove a scratch copy that got shipped, and stop the next one

Reprice the button after a run, and name the encoder correctly

Cross check the rig against the checkpoint that declared it

Make deskate reach the exports, and stop truncating the MCP docs

Correct the licence field, the weight total and the quoted wait

Lay Advanced out as a grid, two rows not four

Document the licence, the retargeting route and the limits

Move Diffusion steps into Advanced, name the download in full

Give BVH its own download again

Keep the label updater out of the MCP tool list

Serve the generator as an MCP tool

Put the wait on the button, not on a slider

Size the two text boxes to their actual jobs

Move the run log under Examples and the backend line to the top

Fix the duplicated Frames slider, and price the steps label

Put Generate below the prompt box

Document the streamed outputs and where BVH went

Stream an elapsed log, and six UI corrections

Document that a dropped stream is terminal

Document the SSE contract and the local ffmpeg gap

Refuse an unknown motion model instead of running SOMA

Quote encoding from the measured 114 s

Mark the encoder executable

Preload the text bundle as one repo entry

Ship real free text, drop the preset embeddings

Refit the estimator to measurements taken after preload

Add a motion model picker for all four checkpoints

Fit the wait estimate to measured Space timings

Single app.py, no header, preload the motion model at build

Snap near-miss prompts to the closest preset

Use measured Space timings for the wait estimate

Drop show_api from launch, removed in Gradio 6

Replace with Kimodo text to motion on CPU via kimodo.cpp

chore: quota lifecycle test, org at exactly limit

Add model size dropdown (small/large), lazy-load LLM sessions, infer KV geometry from ONNX shapes

fix int32->int64 for LLM inputs, fix logits shape indexing

add logic test script

use gradio 6.3.0, drop deprecated type=numpy on Audio output

fix piano key positions, Svelte event dispatch, black key gap

add magenta RT2 piano CPU space