Generative AI

AI

I Had Codex Turn My Illustration Into a ChatGPT Desktop Pet

One illustration plus a two-line request: Codex redrew my character as a chibi, generated 9 animations with subagents, then set it up as a ChatGPT desktop pet.
Claude

Claude Code "Input Is Too Long" and /compact Fails? Rewind With Esc

When every message returns "Input is too long" and even /compact fails, press Esc twice to rewind before the overflow. Why auto-compact misses it, plus prevention.
Claude

Claude Code Stopped With [reasoning_extraction]? Start a New Session

Claude Code flagged my message with [reasoning_extraction] and stopped. Rewind, /model, even loading it into another session failed. What it means, what worked.
Claude

'claude' Is Not Recognized on Windows? Fix PATH for Claude Code, Codex

Installed Claude Code or Codex on Windows but get 'not recognized'? Mixed-up PowerShell/CMD commands, a PATH the installer doesn't set, or an old window. Tested fixes.
Claude

Claude Code Switched From Opus 5.5 to Opus 4.8? Why, and What to Do

Opus 5.5 safeguards flagged my session as [cyber] and switched it to Opus 4.8. How the classifier works, how to switch back with /model, and what kept it from recurring.
Automation

Local AI Video on an 8GB GPU: Claude Code Built It, 16x Faster

I let Claude Code set up local AI video (ComfyUI + Wan 2.2) on an 8GB RTX 2080 SUPER. Switching to tiled VAE decode cut a video from 20 minutes to about 78 seconds.
Generative AI

SDXL on 8GB VRAM: --medvram Cut Generation from 18 Min to Under 1

I removed --xformers, --medvram and --no-half-vae one at a time in AUTOMATIC1111 with SDXL on an 8GB GPU. Without --medvram it silently took 18 minutes. Keep all three.
Generative AI

Wan 2.2 Image-to-Video on 8GB VRAM: Best ComfyUI Settings (10 Min)

Tested settings for Wan 2.2 image-to-video in ComfyUI on an 8GB GPU: Q8_0 model, 14 steps and --fast make a clean 5-second video in about 10 minutes. Workflow included.
Generative AI

Free VRAM Used by Ollama and AivisSpeech for Image Generation (8GB GPU)

Idle Ollama models and GPU text-to-speech quietly hold 4GB+ of VRAM. Unload Ollama with ollama stop and run AivisSpeech on the CPU; measured timings on an 8GB GPU.
Generative AI

Free Local AI Character Chat with Voice and Pictures on 8GB VRAM

Chat out loud with an AI character that also draws the scene, all local and free: Ollama, AivisSpeech and ComfyUI tied together by one PowerShell script, on 8GB VRAM.
スポンサーリンク