Tags give the ability to mark specific points in history as being important
-
-
-
-
-
-
-
-
-
v0.3.8
8b53d948 · ·v0.3.8 — skill security documentation hardening (guardrails at every destructive-endpoint mention)
-
-
-
-
v0.3.4
09356621 · ·v0.3.4 — @psyb0t/flickies ClawHub plugin (MCP bridge) + unified skill+plugin publish
-
v0.3.3
889a5b9e · ·v0.3.3 — re-tag current HEAD to publish the new agent skill to ClawHub. The prior release built fine; only the skill-publish job needed a fresh run to pick up the reusable-workflow fix (version-check no longer aborts on a new skill's 404).
-
-
v0.3.1
d0c1395c · ·flickies v0.3.1 — DEBUG observability pass + engine-unload memory-leak fix Fixed: - Engine unload (wav2lip / gfpgan / latentsync) now frees GPU/CPU memory. Added gc.collect() before torch.cuda.empty_cache() so the cyclic model graph is actually reclaimed — previously weights stayed resident after eviction. Applies to idle sweep, DELETE /v1/engines/{slug}, and hot-swap. Added: - Opt-in DEBUG tracing (FLICKIES_LOG_LEVEL=DEBUG): ffmpeg commands + results, transform decisions (stream_copy vs precise_reencode), inference wall_secs, URL byte counts, job lifecycle, HTTP 4xx/5xx rejections. Security: - Logged URLs are query-stripped so presigned credentials never hit the logs. No API surface change. Backwards-compatible. Spec info.version -> 0.3.1. -
v0.3.0
89e16fd0 · ·flickies v0.3.0 — frame-accurate `precise` flag on trim + concat New optional `precise: bool` (default `false`) on: POST /v1/video/trim POST /v1/video/concat (plus the matching MCP tools) When false: keeps the existing `-c copy` stream-copy fast path. When true: re-encodes (libx264 CRF 18 / AAC 192k) for frame-accurate boundaries — fixes start_sec snapping to the nearest keyframe and fixes concat of inputs with mismatched codec/timebase/SAR. Defaults preserve prior behavior exactly. Pure additive change. Spec info.version bumped to 0.3.0; Pydantic + Python clients regenerated.
-
v0.2.0
20d13549 · ·flickies v0.2.0 operational hardening — structured JSON logging w/ redactor + scope propagation, full HF cache reusable layout, boot-time weight prefetch, X-Request-Id shape validation, bearer auth verified live Live-verified on RTX 3060 12 GB (this turn): wav2lip 23s, gfpgan 83s, latentsync-1.5 477s. Each X-Request-Id correctly threaded through every component log line in the request lifecycle. See CHANGELOG.md for the full list.
-
v0.1.0
38c434b2 · ·flickies v0.1.0 initial public release — real ML inference (Wav2Lip / Wav2Lip-GAN / GFPGAN / LatentSync 1.5), MCP /v1/mcp with 11 tools, generated Go + Python clients, uv.lock + hash-locked heavy deps, both Docker images build clean Engines live on RTX 3060 12 GB: - wav2lip CPU 44s / GPU 22s on a 3s clip (~3.1 GB VRAM) - gfpgan GPU 49s on a 6s clip (~4.2 GB VRAM) - latentsync-1.5 GPU 170s on a 6s clip (~9.6 GB VRAM peak) See CHANGELOG.md for the full list.