Ideogram 4 GGUF Workflow — v2.0 - TurboTime (1 UNET)

The Ideogram 4 GGUF Workflow for ComfyUI generates images from a natural-language prompt with Ideogram 4 in GGUF-quantized format, using UNET, text-encoder, and VAE files; Gemma 4 can optionally convert the prompt to JSON.

Choose v1.0 Standard with two UNET GGUF models or v2.0 TurboTime with one UNET GGUF model.

The optional Gemma 4 prompt enhancer converts natural-language prompts to JSON for text rendering, typography, and composition control.

Inputs include a natural-language prompt, a UNET file, a text encoder, and a VAE; the prompt enhancer is optional.

The documented output is a generated image, and the graph includes PreviewAny and SaveImage nodes.

Update ComfyUI before use, then install molbal/ComfyUI-GGUF with the UnetLoaderGGUF and GGUFCLIPLoader nodes.

Suggestion · not verified

Place UNET files in ComfyUI/models/unet/, text encoders in ComfyUI/models/clip/, and the VAE in ComfyUI/models/vae/.

Suggestion · not verified

Named model files include flux2-vae.safetensors, gemma4_e4b_it_fp8_scaled.safetensors, ideogram_4_turbotime_v1.safetensors, ideogram4-Q4_0.gguf, and Qwen3-8B-Q4_K_M.gguf.

Named graph components include BasicGuider, CLIPLoader, CLIPLoaderGGUF, CLIPTextEncode, ComfyMathExpression, EmptyFlux2LatentImage, Ideogram4Scheduler, KSamplerSelect, LoraLoaderModelOnly, MarkdownNote, and PreviewAny.

Additional named graph components include PrimitiveInt, PrimitiveStringMultiline, RandomNoise, ResolutionSelector, SamplerCustomAdvanced, SaveImage, StringConcatenate, TextGenerate, UnetLoaderGGUF, VAEDecode, and VAELoader.

Choose Q4_0 for lower VRAM use, Q5_0 or Q5_1 for a balance, or Q8_0 for higher quality; Q8_0 is slower.

Suggestion · not verified

For TurboTime v2.0, use one ideogram4-transformer-q4_0.gguf with no unconditional model, CFG 0.0, and scheduler mu=0.5, std=1.75; Standard v1.0 uses both main and unconditional models, CFG 7.0, and mu=0.0, std=1.5.

Suggestion · not verified

Check the listed minimums before use: 8GB+ GPU VRAM, 16GB+ RAM, and 32GB+ virtual memory. If VAE decoding crashes, check that the Windows pagefile is at least 32GB.

Suggestion · not verified

Verify that only molbal/ComfyUI-GGUF is installed for the GGUF loader; city96/ComfyUI-GGUF, ComfyUI-GGUF_KREA-2, and other GGUF loaders are listed as possible conflicts.

Suggestion · not verified