aboutsummaryrefslogtreecommitdiffstats
AgeCommit message (Collapse)AuthorFilesLines
2 daysUse resident daemon model on bare prompt runDanilo M.1-6/+3
Daemon-up prompt run with no -m flag showed the interactive model menu defaulting to sdxl, which then swapped away the model the user loaded at 'imggen start'. Now a bare run adopts the resident model (queried from /status); -m KEY remains the explicit override that swaps. Menu removed as redundant. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysAdd TODO with img2img (image editing) idea for future brainstormingDanilo M.1-0/+38
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysDocument ~/bin deploy step for the wrapperDanilo M.1-2/+12
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysValidate dimensions in the daemon /generate pathDanilo M.2-6/+14
The one-shot CLI already rejects non-multiple-of-8 width/height up front, but the daemon passed them straight through, so an odd dimension over HTTP produced a downstream 500 or garbled output instead of a clear error. Add the same %8 check to /generate, returning 400 with a specific message. Also correct the generate.py scope comment: the registry now holds SDXL-architecture fine-tunes, not just SDXL base. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysDocument model menu, daemon lifecycle, and compose setupDanilo M.2-26/+132
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysFix pony repo and surface daemon swap failures (Arc integration)Danilo M.2-3/+11
Integration on the real Arc B580 uncovered two issues: - The pony repo AstraliteHeart/pony-diffusion-v6-xl is gated (HF returns 401 Repository Not Found), so swaps to it failed. Point the pony key at kitty7779/ponyDiffusionV6XL, an ungated diffusers-format mirror with safetensors weights that loads cleanly via the existing fp16 fallback. - The wrapper swallowed a failed model swap: under set -e the unguarded swap curl aborted the script with exit 0 and no message. Capture the /model response and HTTP status, print the daemon's error, and exit 1 when the swap does not return 200. sdxl, realvis, and pony all verified end to end (one-shot + daemon), output owned by the host user, peak within the 12GB budget. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysRewrite wrapper: daemon lifecycle, client, and one-shot fallbackDanilo M.1-23/+121
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysAdd compose service (runs as host user) and .env.exampleDanilo M.3-0/+28
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysIgnore Python bytecodeDanilo M.1-0/+4
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysAdd persistent HTTP daemon holding one model resident in VRAMDanilo M.1-0/+131
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysAdd -m/--model flag to one-shot CLIDanilo M.1-2/+8
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysRefactor build_pipe to take a model key with fp16-variant fallbackDanilo M.1-9/+16
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysAdd model registry and resolve_model with valid-key errorsDanilo M.2-1/+42
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysAdd implementation plan for model menu + daemonDanilo M.1-0/+739
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysAdd width/height flags and correct default host pathsDanilo M.2-5/+25
Pre-plan prep: -W/-H flags on generate.py with multiple-of-8 validation, generate() gains width/height params, copyright line normalized, and the wrapper defaults updated to the real host output/cache paths. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 daysAdd design spec for model menu + persistent daemon + composeDanilo M.2-0/+238
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
4 daysInitial commit: local SDXL image generation on Intel Arc B580Danilo M.5-0/+368
Containerized SDXL-base generation targeting a 12GB Arc B580 under Slackware. The container carries the Intel Level Zero runtime the host distro does not package; the fp16-fix VAE removes the fp32 upcast that would otherwise OOM at decode. - imggen: host wrapper (GPU passthrough, /out bind-mount, model cache) - generate.py: in-container XPU pipeline, scope-bounded to SDXL base - Dockerfile: pinned to the known-good torch-XPU / diffusers stack - README.md, CLAUDE.md: rationale, limits, and version-pinning notes Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>