Storytelling Image Generation on GPT-image quality: medium 1.0
1Success RateMini-Storyteller-LLaMA-3.2-1B-SFT
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Mini-Storyteller-LLaMA-3.2-1B-SFTBase LLM=Llama-3.2, Size=1B, Training Strategy=SFT, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 172.13 | 8.7 | 0.344 | 57.2 | 45.4 | |
| Llama-3.2-3B-InstructBase LLM=Llama-3.2, Size=3B, Training Strategy=Instruct, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 389.77 | 7.8 | 0.35 | 36.3 | 35.6 | |
| Mini-Storyteller-LLaMA-3.2-3B-SFTBase LLM=Llama-3.2, Size=3B, Training Strategy=SFT, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 322.6 | 8.13 | 0.395 | 46 | 41.4 | |
| Mini-Storyteller-LLaMA-3.2-3B-DPOBase LLM=Llama-3.2, Size=3B, Training Strategy=DPO, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 289.13 | 7.27 | 0.399 | 41.5 | 39.3 | |
| Llama-3.1-8B-InstructBase LLM=Llama-3.1, Size=8B, Training Strategy=Instruct, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 359.27 | 7.53 | 0.374 | 34.2 | 32.1 | |
| Mini-Storyteller-LLaMA-3.1-8B-SFTBase LLM=Llama-3.1, Size=8B, Training Strategy=SFT, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 233.7 | 7.97 | 0.389 | 49.4 | 49.4 | |
| Mini-Storyteller-LLaMA-3.1-8B-DPOBase LLM=Llama-3.1, Size=8B, Training Strategy=DPO, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 182.83 | 7.7 | 0.39 | 49.2 | 43.3 | |
| Mini-Storyteller-LLaMA-3.1-8B-DPO-MixBase LLM=Llama-3.1, Size=8B, Training Strategy=DPO-Mix, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 112.43 | 7.3 | 0.395 | 51.4 | 47.2 | |
| Qwen2.5-1.5B-InstructBase LLM=Qwen2.5, Size=1.5B, Training Strategy=Instruct, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 443.2 | 8.13 | 0.407 | 31.1 | 29.9 | |
| Mini-Storyteller-Qwen2.5-1.5B-SFTBase LLM=Qwen2.5, Size=1.5B, Training Strategy=SFT, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 164.23 | 8.33 | 0.334 | 46.3 | 36.8 | |
| Mini-Storyteller-Qwen2.5-1.5B-DPOBase LLM=Qwen2.5, Size=1.5B, Training Strategy=DPO, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 201.17 | 7.83 | 0.418 | 40.4 | 38.6 | |
| Qwen2.5-7B-InstructBase LLM=Qwen2.5, Size=7B, Training Strategy=Instruct, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 165.83 | 7.87 | 0.381 | 37.2 | 37.8 | |
| Mini-Storyteller-Qwen2.5-7B-SFTBase LLM=Qwen2.5, Size=7B, Training Strategy=SFT, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 157.77 | 8.23 | 0.342 | 45.2 | 42.9 | |
| Mini-Storyteller-Qwen2.5-7B-DPOBase LLM=Qwen2.5, Size=7B, Training Strategy=DPO, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 221.43 | 7.73 | 0.394 | 46.3 | 42.4 | |
| Mini-Storyteller-Qwen2.5-7B-DPO-MixBase LLM=Qwen2.5, Size=7B, Training Strategy=DPO-Mix, Painter Model=GPT-image-1 (medium)2025.12 | 1 | 153.63 | 8.47 | 0.39 | 49.5 | 47.1 | |
| Llama-3.2-1B-InstructBase LLM=Llama-3.2, Size=1B, Training Strategy=Instruct, Painter Model=GPT-image-1 (medium)2025.12 | 0.967 | 459.27 | 7.33 | 0.432 | 36.5 | 38.6 | |
| Mini-Storyteller-LLaMA-3.2-1B-DPOBase LLM=Llama-3.2, Size=1B, Training Strategy=DPO, Painter Model=GPT-image-1 (medium)2025.12 | 0.967 | 79.97 | 7.1 | 0.381 | 47.8 | 37.2 |