Ideogram 4 in ComfyUI — FASTEST Setup + The NSFW Secret Nobody's Talking About - Video

Learn how to run Ideogram 4 in ComfyUI at maximum speed using INT8 quantization — a 1.5x–2x speedup optimized for RTX 30xx GPUs. Got a 40xx? Use FP8. On a 50xx Blackwell? You've got full flexibility: NVFP4, MXFP8, whatever you want.
But speed isn't the only thing we're covering. Ideogram 4 has been stirring up controversy, and I'm breaking down exactly how NSFW prompting works in plain natural language — no JSON wrangling, no LLM middleware. Just the Kijai nodes doing the heavy lifting so you can prompt the way you actually think.

Links:
For the workflow, grab it in my Discord under the "ComfyUI-Workflows" channel: https://discord.gg/c45hYQaTjb
To access my INT8 quantized models, go to my Huggingface: https://huggingface.co/Winnougan
For the quantized Ideogram 4 in "convrot" or "tensorwise" formats, smash and grab from Silveroxides: https://huggingface.co/silveroxides/ideogram4-dequant-and-int8-quant
For Triton, Sageattention and Flashattention installations go here: https://github.com/wildminder/AI-windows-whl
Please support any way you can. It goes a long so I can make more AI content like finetune models, LoRAs, videos, custom nodes: https://ko-fi.com/winnougan