| Caption the image in continuous text, covering elements from foreground to background. The caption must be written in English. Refer to the <EXAMPLES> and follow the <STRUCTURE> and <RULES>, <PROHIBIT> below to caption the image. | |
| <RULES> | |
| - The user inputs an image and a background prompt. | |
| - The user input image has a simple, flat background (or the background is removed). The specified background prompt is completed according to these <RULES>. A caption for the character is also required. | |
| - The prompt must be detailed. The prompt must be greater than 700 characters in length. The background and character prompts should be roughly equal in length. | |
| - Write an image caption in a run of text, like the examples in <EXAMPLES>. | |
| - Include the subject, positional relationship of characters, composition, angle of view, character features, character's clothing, character's actions (also write objects necessary for the action), the state of the background, and other objects. | |
| - If parts are unclear due to overexposure (blown-out highlights), underexposure (crushed blacks), or other reasons, write them only if you are more than 50% confident. | |
| - Users can add additional rules and info. Please follow the rules specified by the user. | |
| <PROHIBITS> | |
| - Prohibit past tense and future tense. | |
| - Do not write speculations; write definitively. For example, instead of "floral pattern (probably cherry blossom)," write "floral pattern" or "cherry blossom floral pattern." | |
| - Prohibit including structure item names in the caption from the "<STRUCTURE>" section (e.g., Subject Description, Who, Related culture, etc.). | |
| - Prohibit "probability" prompts that express probability (e.g., seems like, looks like, appears to be, likely, probably, possibly, might, may, could, suggests, implies, indicates, is thought to be, is inferred to be). | |
| - Do not mention any text contained in the image. | |
| - Consider the angle of the user input image, and do not include elements that should not be present within that angle. As a failure example, consider a 'close-up' and 'top-down angle' that nevertheless includes the word 'sky.' However, if it's a 'panoramic view,' then including it might be acceptable. Use your discretion. | |
| - Do not mention simple backgrounds included in the user's input image. You should generate the background based on the user's input prompt. | |
| For example, in the background prompt, "The background is a vibrant, sun-drenched beach with crystal-clear turquoise waters lapping at the shore. White, soft sand stretches out behind him, dotted with a few scattered seashells. Palm trees with lush green fronds sway gently in the distance under a bright blue sky with a few wispy white clouds. The image adopts a full-body composition, viewed from a high angle looking down at the character.", the following part is against the rules: "a bright blue sky with a few wispy white clouds. ... viewed from a high angle looking down at the character." | |
| - Prohibit including words "anime" and "photo". | |
| <EXAMPLES> | |
| - (Example 1) The girl wearing a black dress, her eyes are purple, and she is holding a blue rose in her hand. The background is dark blue. The image is composed in the middle ground. The half body. | |
| - (Example 2) A red-haired man wearing only his underwear, sitting on a rock, his wings behind his back, the background is a beautiful landscape. The image adopts panoramic composition, the whole body. | |
| <STRUCTURE> | |
| Subject Description: | |
| Who: e.g. "The girl" (Example 1), "A red-haired man" (Example 2), | |
| Attire: e.g. "wearing a black dress" (Example 1), "wearing only his underwear" (Example 2), | |
| Physical features/Items: e.g. "her eyes are purple, and she is holding a blue rose in her hand" (Example 1), "his wings behind his back" (Example 2), | |
| Action/Pose: e.g. "sitting on a rock" (Example 2), | |
| Background Description: | |
| What kind of background: e.g. "The background is dark blue." (Example 1), "the background is a beautiful landscape" (Example 2), | |
| Composition and Shot Description: | |
| Type of composition: e.g. "composed in the middle ground" (Example 1), "adopts panoramic composition" (Example 2), | |
| Shot range: e.g. "The half body" (Example 1), "the whole body" (Example 2), |