<span class="emp-dg"> SoloTagger is an LLM-based image captioning tool. </span>
## What's Changed
1. The prompt passing mechanism has been optimized, **significantly improving caption output quality**.
2. The default **temperature** has been changed from **0.1 to 0**, reducing model hallucinations.
## **Note:**
1. The prompt is the most important factor affecting caption output quality. Please refine and adjust it according to your specific needs.
2. In my own experience, with a well-crafted prompt, <span class="emp-y">Qwen3.5-9B-UD-Q5_K_XL.gguf</span> produces better output quality than <span class="emp-y">llama-joycaption-beta-one-hf-llava.i1-Q6_K.gguf</span>.
3. Among the **Qwen3.5** series, models smaller than **9B** produce noticeably lower output quality.
For detailed usage instructions, please refer to the previous documentation:
[v0.12: Easier Prompt Editing](https://sololo.xyz/article/25-solotagger-v012-easier-prompt-editing)
[v0.10: Setup on Windows via LM Studio](https://sololo.xyz/article/24-solotagger-local-joycaption-beta-one-gguf-setup-on-windows-via-lm-studio)
**Download:** <a href="https://assets.sololo.xyz/articles/026/SoloTagger_v0.20.zip">SoloTagger_v0.20.zip</a>
**GitHub:** https://github.com/sololo-xyz/SoloTagger
