ltx 2.5 gemma-4-E2B-it Unexpected text model architecture type

[kpu3uc]

https://huggingface.co/llmfan46/gemma-4-E2B-it-ultra-uncensored-heretic-GGUF/blob/main/gemma-4-E2B-it-ultra-uncensored-heretic-BF16.gguf
for Generate LTX2 Prompt node.
ValueError: Unexpected text model architecture type in GGUF file: 'gemma4'

[kpu3uc]

[kpu3uc]

[ERROR] !!! Exception during processing !!! Unexpected text model architecture type in GGUF file: 'gemma4'
[ERROR] Traceback (most recent call last):
File "F:\3\ComfyUI\execution.py", line 545, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\3\ComfyUI\execution.py", line 344, in get_output_data
return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\3\ComfyUI\execution.py", line 318, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "F:\3\ComfyUI\execution.py", line 306, in process_inputs
result = f(**inputs)
File "F:\3\ComfyUI\custom_nodes\ComfyUI-GGUF\nodes.py", line 251, in load_clip
return (self.load_patcher([clip_path], clip_type, self.load_data([clip_path])),)
~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "F:\3\ComfyUI\custom_nodes\ComfyUI-GGUF\nodes.py", line 227, in load_data
sd = gguf_clip_loader(p)
File "F:\3\ComfyUI\custom_nodes\ComfyUI-GGUF\loader.py", line 471, in gguf_clip_loader
sd, extra = gguf_sd_loader(path, is_text_model=True)
~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\3\ComfyUI\custom_nodes\ComfyUI-GGUF\loader.py", line 108, in gguf_sd_loader
raise ValueError(f"Unexpected text model architecture type in GGUF file: {arch_str!r}")
ValueError: Unexpected text model architecture type in GGUF file: 'gemma4'

[Leavitskiy]

Worth separating two things that both say "gemma" here: the node you're
loading into (Generate LTX2 Prompt) is the prompt enhancer — a separate LLM
that rewrites your prompt — not the text encoder that conditions the
diffusion. Those are different models, gemma4_e2b_it (~9.6 GiB) vs
gemma4-12b-with-proj-ltx-2.5 (~14.3 GiB with LTX's own projections).

The enhancer is optional and ships enabled in every LTX-2.5 template, which
catches people out. If you're doing this to save VRAM, turning it off frees
the whole ~9.6 GiB and costs you nothing but auto-rewritten prompts — write
them yourself or in any chat model. That's likely a better outcome than a
quantized enhancer even if the arch were supported.

(Doesn't address the actual gemma4 arch gap in the loader, which is a real
request — just that it may not be the thing you need.)