WikiQwen 4B
Generate illustrated how‑to articles from any question
Ask anything, get an illustrated how-to article: chat models, an illustration LoRA and the demo.
Generate illustrated how‑to articles from any question
Note Try it: ask anything, get a step-by-step article with a picture for every step.
Generate illustrated step‑by‑step how‑to articles
Note WikiQwen-4B writes, WikiQwen-Illustrator-9B draws: sharper pictures, same small GPU.
Generate illustrated step-by-step how‑to articles from a question
Note WikiQwen-9B writes, WikiQwen-Illustrator-9B draws: the strongest pairing (best for logged-in users).
Note Qwen3.5-4B fine-tuned to answer everything as an illustrated how-to article.
Note Qwen3.5-9B fine-tuned the same way: more knowledgeable, a little slower than the 4B.
Note Block-wise FP8 build of WikiQwen-9B: 11 GB, held-out loss +0.26%.
Note The largest writer: Qwen3.6-35B-A3B mixture of experts, about 3B active per token.
Note FP8 experts build of WikiQwen-35B-A3B: 37 GB, fits on one GPU.
Note Style LoRA for FLUX.2 [klein] base 4B that draws each step's caption.
Note Style LoRA for FLUX.2 [klein] base 9B (FLUX Non-Commercial License): better hands and objects.