Media Gen Roadmap
Evaluation of image / video / audio generation for this machine. The same constraints
that shaped the LLM picks apply here — but media workloads invert the memory story:
diffusion is bandwidth-bound rather than token-bound, so the 256 GB/s LPDDR5X
(measured ~212 GB/s) is the ceiling, not the 124 GiB GTT pool. Full findings:
research/mediagen-eval.md.
The one hardware fact that decides everything
Diffusion video/image generation is memory-bandwidth-bound. Measured community scaling on Wan 2.2 vs an RTX 3090 (936 GB/s): 320×320 lands at ~36% of the 3090, 512×512 at ~19%, 640×640 at ~11%. That collapse is an unresolved gfx1151 ROCm kernel inefficiency (TheRock #2591) on top of the raw bandwidth gap. Video works on this box; it is not fast, and it degrades hard above 640px.
The genuine hardware win is fit: no consumer NVIDIA card holds unquantized Wan 2.2 14B. This machine does — that's a batch-mode capability, not a speed one.
Selected picks — by modality
| Modality | Pick | Runtime | Footprint | Expected perf | License |
|---|---|---|---|---|---|
| Images (daily) | Z-Image-Turbo 6B, GGUF Q8 + Qwen3-4B encoder | stable-diffusion.cpp v1.0.1 (Vulkan) | ~12 GB total | ~10 s/image (AMD-confirmed on Strix Halo) | Apache 2.0 ✓ |
| Images (text-heavy / edits) | Qwen-Image-2512 20B, GGUF Q8 | ComfyUI + TheRock wheels | ~21 GB (Q8) | ~120–150 s/image | Apache 2.0 ✓ |
| Video (entry) | Wan 2.2 TI2V-5B + Lightning LoRA | ComfyUI | fits unquantized | ~8.5–9 min / 5 s 720p clip | Apache 2.0 ✓ |
| Video (batch) | Wan 2.2 14B + 4-step Lightning | ComfyUI | 54–80 GB unquantized | ~30 min / 832×480 clip | Apache 2.0 ✓ |
| TTS | Kokoro-82M | CPU (no GPU needed) | <1 GB | faster than real-time on CPU | Apache 2.0 ✓ |
| Voice cloning | Chatterbox (v3 Multilingual) | GPU | ~4–6 GB VRAM | real-time-ish | MIT ✓ (PerTh watermark default-on) |
| STT | whisper large-v3-turbo | whisper.cpp Vulkan | ~1.6 GB | ~26× real-time (Windows Vulkan datapoint) | MIT ✓ |
Considered and excluded — with reasons
- Flux.2 [dev] 32B — best quality available, but BFL non-commercial license for the weights (paid license required in a commercial product) and ~32 GB at Q8. ComfyUI-only; stable-diffusion.cpp has no Flux.2 support yet. License-excluded for any monetized path.
- Flux.1 [dev] — same BFL non-commercial weights; superseded by Z-Image/Qwen-2512 on quality-per-watt anyway. Personal/lab use only.
- F5-TTS / CosyVoice (cloning) — look open, but both carry CC-BY-NC-4.0 model weights (Emilia / WenetSpeech4TTS training data). Banned for monetized use. Chatterbox is the only SOTA-quality cloner with permissive code and weights.
- HunyuanVideo — 35–50 min per 5 s clip on this bandwidth class; strictly worse than Wan on this hardware.
- MiniMax H3 (video+audio joint DiT, Apache 2.0, merged into ComfyUI core Aug 2026) — most active Sep-2026 ecosystem, 19.6–31.7 GB INT8 fits easily, but no Strix Halo perf datapoint yet. Watch-list.
- Meta Muse Image — API-only, no local weights. Out of scope for an on-box build.
- SD.Next / FastSDCPU — smaller communities, fewer gfx1151 datapoints than sd.cpp or the kyuz0 ComfyUI toolboxes. Not worth the deviation.
Co-residency rule vs the LLM pool
The 124 GiB GTT pool is already claimed by LLM serving (up to 96–113 GB resident). Only three media workloads are exempt from swap-out:
- Kokoro (~1–3 GB) and whisper turbo (~1.6 GB) — always resident, zero conflict.
- Z-Image-Turbo (~12 GB with encoder) — co-resident whenever ~15 GB of GTT headroom exists (fine alongside Qwen3.8-27B; conflicts only with the 100 GB+ MoE models resident).
Everything else (Qwen-Image full precision, Wan 14B, Flux.2) requires unloading the served LLMs first. Rule: any media job >16 GB swaps the LLMs out and restores after. Treat video as an overnight batch mode, not an interactive one.
Install-order plan (draft — nothing executed)
- Phase 0 — zero risk, co-resident: whisper.cpp Vulkan build + large-v3-turbo; Kokoro CPU service. Verify with a 10-min transcription RTF and a 5-min narration render.
- Phase 1 — low risk: stable-diffusion.cpp v1.0.1 Vulkan build; Z-Image-Turbo GGUF Q8 (7.2 GB) + Qwen3-4B encoder. Bench target ~10 s/image.
- Phase 2 — medium complexity, LLM swap required: ComfyUI via the
kyuz0 Fedora toolbox (closest match to Fedora 44); TheRock multi-arch wheels with
[device-gfx1151]extras; the established env set (HSA_OVERRIDE_GFX_VERSION=11.5.1,HSA_ENABLE_SDMA=0,HSA_USE_SVM=0,--disable-mmap,attention_mode=sdpa,cudnn.enabled=false). Then Qwen-Image-2512 Q8 and Wan 2.2 TI2V-5B. - Phase 3 — on demand: Chatterbox for voice cloning, watermark-on.
Biggest risk per modality
- Images: Qwen-Image-2512 produces black images on k-quant GGUFs (sd.cpp issue #1385, activation dequant overflow) — stay on Q8 and lead with Z-Image.
- Video: the gfx1151 ROCm kernel inefficiency (TheRock #2591) has no fix timeline; budget 10–30 min/clip and cap resolution at 480p for the 14B.
- Audio: the cloning-license trap above; and Chatterbox embeds a PerTh neural watermark in every output by default — a disclosure fact, not a defect.