Best AI for character reference sheets
The short answer
FLUX 3 Image and Grok Image 2.0 gave us a complete, consistent sheet on all 10 attempts: three views and four expressions, with the same outfit in every view. GPT Image 2.5 slipped twice (a sword missing from the back view, an extra view) but is the cheapest at 1 credit. Ideogram 4.5 failed most often: close-ups pasted over the views, and once a bald head in the expression row.
Results
Two attempts per model per brief, no cherry-picking. A sheet passes when it has the front, side and back views and all four expressions, with the same character, outfit and props in every view and no text. A missing prop or one extra or missing panel is a small slip.
| Model | Attempts that pass our checks | Credits per sheet (2K) | Credits per usable result |
|---|---|---|---|
| Grok Image 2.0xAI · medium quality, 2K | 10 / 10reliability 100% | 2.5 | 2.5 |
| FLUX 3 ImageBlack Forest Labs · 2K | 10 / 10reliability 100% | 3 | 3 |
| GPT Image 2.5OpenAI · Sunburst, medium quality, 2K | 8 / 10reliability 90% | 1 | 1 |
| Nano Banana 2.1Google · 2K | 7 / 10reliability 85% | 2 | 2 |
| Seedream 5.0 ProByteDance · 2K | 6 / 10reliability 80% | 2.5 | 2.5 |
| Recraft V4.1Recraft · standard, 2K | 3 / 10reliability 65% | 8 | 8 |
| Ideogram 4.5Ideogram · medium quality, magic prompt off, 2K | 2 / 10reliability 45% | 2 | 2.86 |
The blind panel is grading these outputs now. BiBi Scores, which add the panel's looks score and adjust for each grader's habits, appear here when grading finishes. How scoring works.
Every output
Unedited, both attempts, in the order the models were run. Click an image to see it full size.
Delivery rider (realistic)
PromptCharacter reference sheet on a plain white background, 16:9: a woman in her mid-twenties, East Asian, a food-delivery rider in a bright yellow jacket with a reflective stripe, black trousers and white sneakers, holding a yellow helmet. Show the full body from the front, from the side and from the back, plus four head close-ups with expressions: neutral, smiling, surprised, tired. The same person, clothes and details in every view. Photorealistic. No text or labels.
Retired jazz musician (realistic)
PromptCharacter reference sheet on a plain white background, 16:9: an elderly Black man in his seventies, a retired jazz musician with a short grey beard, a brown flat cap, round glasses, a mustard cardigan over a white shirt and dark brown trousers. Show the full body from the front, from the side and from the back, plus four head close-ups with expressions: neutral, smiling, surprised, tired. The same person, clothes and details in every view. Photorealistic. No text or labels.
Astronaut (realistic)
PromptCharacter reference sheet on a plain white background, 16:9: a South Asian woman in her thirties, an astronaut in a white spacesuit with an orange shoulder patch, helmet off and held under one arm, short black hair. Show the full body from the front, from the side and from the back, plus four head close-ups with expressions: neutral, smiling, surprised, tired. The same person, suit and details in every view. Photorealistic. No text or labels.
Wuxia swordsman (anime)
PromptAnime-style character reference sheet on a plain white background, 16:9: a young Chinese wuxia swordsman in a flowing blue robe with a white sash, black hair tied in a topknot, a sword in a dark scabbard at his hip. Show the full body from the front, from the side and from the back, plus four head close-ups with expressions: neutral, smiling, surprised, angry. The same character, outfit and details in every view. Clean 2D anime style. No text or labels.
Chef cat (3D cartoon)
Prompt3D cartoon character reference sheet on a plain white background, 16:9: a round, chubby orange cat who is a chef, wearing a tall white chef's hat, a white double-breasted jacket and a small red neckerchief. Show the full body from the front, from the side and from the back, plus four head close-ups with expressions: neutral, happy, surprised, grumpy. The same character and details in every view. Polished 3D animated-film style. No text or labels.
How we tested
Five characters: three realistic people, one anime swordsman and one 3D cartoon cat. The prompt asked for a 16:9 sheet on white with three full-body views and four head close-ups with named expressions, and no text. Every model ran at 2K, twice per character, on Higgsfield on 3 Nov 2026.
We checked every sheet for missing or extra views, outfit or prop changes between views, and labels. Our blind panel then grades whether a video team could use the sheet as a reference.