---

base_model: lightx2v/MiniMax-H3-Prompt-Rewriter-LoRA-8B

base_model_relation: adapter

library_name: gguf

pipeline_tag: image-text-to-text

tags:

- gguf

- lora

- llama.cpp

- qwen3-vl

- multimodal

- prompt-rewriting

- minimax-h3

- text-to-audio-video

- image-to-audio-video

language:

- en

---


MiniMax-H3 Prompt Rewriter LoRA 8B — GGUF

A GGUF conversion of

lightx2v/MiniMax-H3-Prompt-Rewriter-LoRA-8B,

so the multimodal rewriter runs under

llama.cpp against a quantized

Qwen3-VL-8B-Instruct.

This repository contains no new training — only a format conversion. The tensor

values are unchanged.

What is different about this one

The 27B rewriter

reads text and covers one task. This adapter's base model has a vision tower, so

it looks at the reference frames rather than being told about them in words,

and it covers four tasks:

| Task | Pictures | Meaning |

|---|---|---|

| T2AV | 0 | text only |

| I2AV | 1 | the picture is the first frame |

| L2AV | 1 | the picture is the final frame |

| FL2AV | 2 | first and last frame |

For everything but T2AV the answer opens with an alignment sentence that

MiniMax-H3 itself reads, with the timestamps already filled in.

It is also much smaller than the 27B route:

| Base | Download | VRAM with this adapter |

|---|---|---|

| Qwen3VL-8B-Instruct-Q4_K_M.gguf + projector | 4.7 + 0.7 GB | ~9 GB |

| Qwen3VL-8B-Instruct-Q8_0.gguf + projector | 8.1 + 0.7 GB | ~13 GB |

A multimodal model is two files, and the projector has to come from the same

conversion as the base. Both are in

Qwen/Qwen3-VL-8B-Instruct-GGUF;

mmproj-Qwen3VL-8B-Instruct-Q8_0.gguf is the smaller of the two projectors and

is what the numbers above assume.

Contents

| File | Size | Notes |

|---|---|---|

| MiniMax-H3-Prompt-Rewriter-LoRA-8B-F16.gguf | 1.30 GB | 504 tensors, rank 256, adapter.lora.alpha = 256 |

| MiniMax-H3-Prompt-Rewriter-LoRA-8B-Q8_0.gguf | 0.69 GB | the same adapter, quantised |

Tensor count and alpha match the source adapter exactly. The adapter touches the

text tower only — q/k/v/o and gate/up/down across 36 layers; the vision tower

is untouched, which is why 504 tensors cover the whole thing.

Take Q8_0 unless you have a reason not to. Names, order and every metadata

field are identical to the F16; only the stored values differ, by a mean of

0.005 relative across all 504 tensors and 0.006 on the worst one. Both produce

the trained output shape: the three fields, [Shot 2] At 00:04.500 cut markers,

(S1) speaker ids and <d>[English] ...</d> dialogue tags.

There is no Q4_K_M. llama-quantize treats an adapter GGUF as a model and

stops at key not found in model: qwen3vl.context_length, so F16 and Q8_0

are the only two formats the tooling offers for a LoRA.

Use with llama.cpp

A task with pictures goes through llama-mtmd-cli, and <__media__> is where

each one is spliced in. The number of markers must equal the number of --image

arguments, or tokenization is refused.

llama-mtmd-cli \
  -m Qwen3VL-8B-Instruct-Q4_K_M.gguf \
  --mmproj mmproj-Qwen3VL-8B-Instruct-Q8_0.gguf \
  --lora MiniMax-H3-Prompt-Rewriter-LoRA-8B-Q8_0.gguf \
  --image first.png \
  -sys "$(cat system_prompt.txt)" \
  -p "Picture 1 — exact first frame at 0.00 seconds:
<__media__>
task: i2av
resolution: 16:9
duration: 10s
original_prompt: A lone astronaut walks toward a ruined tower on a red desert plain." \
  -n 1400 --temp 0 -ngl 99 -c 8192

T2AV has no pictures, so it needs no projector and runs under plain

llama-completion:

llama-completion \
  -m Qwen3VL-8B-Instruct-Q4_K_M.gguf \
  --lora MiniMax-H3-Prompt-Rewriter-LoRA-8B-Q8_0.gguf \
  -sysf system_prompt.txt \
  -p "task: t2av
resolution: 16:9
duration: 10s
original_prompt: A lone astronaut walks toward a ruined tower on a red desert plain." \
  -no-cnv -st -n 1400 --temp 0 -ngl 99 -c 8192

system_prompt.txt must hold the exact system prompt from

prompt_template.py

in the source repository — the adapter was trained on that wording, and changing

it degrades the rewrite. That file also builds the user turn, including which

line introduces each picture, which differs per task.

-st on the second command is worth keeping: llama-cli enters chat mode on

this model despite -no-cnv and then waits for a second turn, so the run never

ends.

Use in ComfyUI

MiniMax-H3-Prompt-Rewriter-ComfyUI

ships a node for this adapter — MiniMax-H3 Prompt Rewriter 8B (sees frames).

Pick a Qwen3-VL base, choose the task, connect first_frame and last_frame

where the task wants them, and this adapter is fetched from here on first use.

The options node's adapter dropdown lists both precisions.

What to expect of it

All four tasks produce the trained shape, and with the adapter off the same base

model stops writing [Shot 2] cut markers and answers about a third as long —

which is the simplest confirmation that it is attached.

It is an 8B, and that shows in one place: the alignment line's timestamp is

sometimes formatted to three decimal places instead of two, and on FL2AV the

final picture is occasionally credited to Shot 1 rather than the last shot. The

27B adapter does not do this.

How it was converted

convert_lora_to_gguf.py from llama.cpp, with nothing done to it. Qwen3-VL is an

ordinary attention architecture, so the V-head reordering that makes the

27B conversion

awkward never comes up here. Both files are separate runs of the converter, not

one quantised from the other.

python convert_lora_to_gguf.py lora8b --outtype f16
python convert_lora_to_gguf.py lora8b --outtype q8_0

No --base is needed: the adapter's own adapter_config.json names

Qwen/Qwen3-VL-8B-Instruct, and the converter reads the base config from the

hub. Only the config is fetched — the base weights are not required.

Licence

The weights originate from

lightx2v/MiniMax-H3-Prompt-Rewriter-LoRA-8B

and were converted without modifying their values. Whatever terms that

repository carries apply here unchanged; this repository claims nothing

additional.

Use of MiniMax-H3 itself is governed by the licence and acceptable-use terms in

the official MiniMax-H3 repository.