[Issue] - Ideogram isn't loading properly.

[JohnLoveJoy]

The model does not load.
Exactly as shown below.

"ImageGen Init - Load Model: E:\IDEOGRAN4\ideogram4_uncond-Q4_0.gguf
With Custom T5-XXL Model: E:\IDEOGRAN4\Qwen3-VL-8B-Instruct-Q3_K_S.gguf
Conv2D Direct for VAE model is enabled

Swap to Diffusion Model Path

:E:\IDEOGRAN4\ideogram4_uncond-Q4_0.ggufSelecting CPU as CLIP image generation device

Try read vocab from C:\Users\JDS\AppData\Local\Temp_MEI69082/embd_res/qwen2_merges_utf8_c_str.embd

Error: Image generation failed to setup!
Make sure you have ALL files required (e.g. VAE, T5, Clip...) or baked into the model!
Load Image Model OK: False

Error: Could not load image model: E:\IDEOGRAN4\ideogram4_uncond-Q4_0.gguf"

[henk717]

Are you on the latest koboldcpp?

[JohnLoveJoy]

Are you on the latest koboldcpp?

Yes. I think I've found the problem... According to the template, encoder(LLM) and unconditional model need to be placed in odd, unintuitive fields like "clip".

[wbruna]

ImageGen Init - Load Model: E:\IDEOGRAN4\ideogram4_uncond-Q4_0.gguf
With Custom T5-XXL Model: E:\IDEOGRAN4\Qwen3-VL-8B-Instruct-Q3_K_S.gguf
Conv2D Direct for VAE model is enabled

Looks like you are missing both the main model and the VAE. On my working config, I get:

ImageGen Init - Load Model: ideogram4-Q4_0.gguf
With Custom VAE: flux2.full_encoder_small_decoder.safetensors
With Custom Clip-1 Model: Qwen3-VL-8B-Instruct-UD-Q4_K_XL.gguf
With Custom Clip-2 Model: ideogram4_uncond-Q4_0.gguf

I suggest loading the .kcppt from the release notes, and changing it to point to the downloaded models.

[LostRuins]

The field are weirdly named but thats because they are overused for all types of architectures. If i added them today I would probably just call it --sdextramodel1 or something.