Replying to @wildmindai
Since when safetensors can be flagged as unsafe ^^
Replying to @jauneclaude
QWEN 3.6 35b vs GEMMA4 27B , same results across multiple runs on similar tasks, gemma4 have better vision
1
2
254
Replying to @nanbeige
Consumes small amount of vram, properly follow the instructions. Perfect to refine or create prompts in a ComfyUI workflow since it's fast to load/unload due to the model size. Got arount 140t/s with a 4090, we need MTP to be faster, could reach 200t/s with it maybe more
3
322
Replying to @Viggle_PINOC
Hymotion is still the best text to rig I tried youtu.be/6y8waqlKbPE?t=117
1
3
245
Replying to @uniqornate @Machinedelusion
I didn't saved any logs from AI toolkit, but if I paid 53$ at 2$/h I trained for about 24hours or so.
But if you want to train a character, consider using 30 good images, and few videos like 5 - 10 (optional), you really don't need more than that for a LoRA
1
1
25
Replying to @mathmpv @MemeCryptoFR
C'est pas un ancien ledger, les angles sont saillants et l'écran plus fin sur les anciens. Dans tout les cas impossible de faire quoi que ce soit avec sans la seed
1
3,697






