Session 1 — 2026-06-07 (Cron Run)
What was done
Kanban setup
- Created kanban card
t_e52a7709: “Synthetic Avatar Pipeline (Mick)” - Status: ready → in progress for this session
Audit of existing work
The research phase is complete (done on 2026-06-05). Key findings:
| Item | Status | Details |
|---|---|---|
| Model selected | ✅ Done | SoulX-FlashHead (Apache 2.0, 96 FPS on RTX 4090) |
| Research doc | ✅ Done | research.md — full model comparison matrix written |
| Reference photos | ✅ Available | 14 PNGs in /opt/data/creative/faces/png/ (IMG_7985–IMG_7999) |
| Dockerfile | ✅ Written | CUDA 12.4, Torch + Diffusers + SoulX-FlashHead from source |
| Inference server | ✅ Written | FastAPI POST /animate, audio + ref image → WebM stream |
| K8s manifests | ✅ Written | Deployment (GPU nodeSelector), Service, Ingress at flashhead.hermes.paralla.org |
Integration design confirmed
Text → F5-TTS (.193) → Audio WAV → SoulX-FlashHead (.106, RTX 4090) → WebM Video Stream
↑
Reference Image (from /opt/data/creative/faces/png/)
What has NOT been done (Phase 2 gaps)
- No inference test — model installed in Dockerfile but never tested end-to-end
- Best reference photo not selected — need visual assessment to pick the best of 14 candidates
- GPU K8s node registration — verify
.106hasnvidia.com/gpuresource - Audio format compatibility — confirm F5-TTS output matches SoulX-FlashHead expected input (likely 16kHz mono WAV)
- Docker image not built/pushed to registry
Blocker for Phase 2
Cannot proceed with actual inference testing without:
- SSH access to .106 or kubectl exec into GPU pod
- Visual inspection capability to select reference photo
- Docker build environment with CUDA/registry access
Next action required: pvs needs to approve Phase 2 kickoff, then either:
- Schedule interactive session to test inference on .106
- Delegate to a worker that has SSH/GPU node access
- Provide credentials for remote testing
Session log summary
| Timestamp | Action |
|---|---|
| 07:20 UTC | Kanban card t_e52a7709 created, status=ready |
| 07:21 UTC | Audit of existing research/infrastructure completed |
| 07:21 UTC | Phase transition documented — ready for implementation |
| 07:21 UTC | Session file written |