CIVITAI / Workflows

Ideogram 4 + KJ Prompt Builder Visual Composition Director Workflow

Watch the full video first if you want to understand how this Ideogram 4 + KJ Prompt Builder workflow works in practice. The video shows how structured JSON prompts can be used to control poster layout, typography, composition, color palette, object placement, and overall visual direction inside ComfyUI. This ComfyUI workflow is designed for Ideogram 4 visual composition control using KJ Prompt Builder. Its main purpose is not only to generate a beautiful image, but to make the image generation process more like a visual design system. Instead of writing one loose natural-language prompt, this workflow uses structured JSON-style prompting and a visual composition builder to define background, subject elements, text blocks, bounding boxes, color palettes, style direction, lighting, and output ratio. The workflow is built around ideogram4_fp8_scaled.safetensors as the main Ideogram 4 model. It also uses ideogram4_unconditional_fp8_scaled.safetensors as the unconditional model branch, qwen3vl_8b_fp8_scaled.safetensors as the Ideogram 4 text encoder, flux2-vae.safetensors as the VAE, Ideogram4Scheduler, DualModelGuider, CFGOverride, EmptyFlux2LatentImage, SamplerCustomAdvanced, VAEDecode, SaveImage, and multiple Ideogram4PromptBuilderKJ nodes. The most important part of this workflow is Ideogram4PromptBuilderKJ. This node allows the user to organize a visual prompt into a more controllable layout format. You can describe the global image concept, background, style, lighting, medium, color palette, and individual elements. Each element can be treated as an object or text block, with its own position, description, text content, and palette. This is especially useful for Chinese posters, thumbnails, title images, brand visuals, concept art, and design-heavy images where placement matters. The workflow also includes multiple prepared example groups, such as Chinese movie poster layouts, racing posters, realistic documentary-style scenes, Chinese fantasy worldbuilding, puzzle-adventure environments, F1 racing, MotoGP racing, gaming showcase visuals, action-horror survival scenes, and dystopian underground sci-fi compositions. These examples make the workflow more than a single generation graph; it becomes a practical visual prompt library for learning how structured Ideogram 4 prompting works. Another important part is the dual-model guidance structure. The main