Seedream의 애니메이션 스타일 이미지 생성 불량을 개선하는 3가지 방법

The previous article is .

In the previous article, I succeeded in verifying the hypothesis for creating beautiful anime-style images with Seedream, but the problem of poor generation of clothing depictions remained.

So, I received various advice in the comments, and based on that, I’m trying again to create beautiful images with Seedream👍
(Thank you to everyone who commented!)

【Three Improvement Strategies】

First, why did the poor clothing generation occur? I hypothesized that the cause was in the way the prompt was written, and I narrowed down the improvement target to this.

Based on the suggestion from 日本及大和號戰艦的应援会™:
"For the costume, start with a simple design, then gradually add decorations like lace and frills for comparison"
and the observation from こばと🕊️_ダヴ(DOVE)さん:
"Prompts written as bullet points are easier to understand"
I also came up with my own improvements and decided to try the following three methods!

① Generate with a simple prompt that does not specify clothing decorations (information reduction approach) ⬇️
② Generate with a prompt rewritten in bullet point format (format change approach) ✂️
③ Generate with a prompt that adds elements to concretize the clothing shape (information addition approach) ⬆️

Incidentally, ③ is an application of the tip I previously wrote about using Image V2: "Adding conspicuous items like hats to stabilize the composition" ⚖️👒
(Reference is .)

This time, I’m experimenting by modifying the part of the previous prompt: "A gothic lolita-style costume in blue and black with a blue ribbon tie, lace and frills on the collar, hem, and sleeves"!

(Also, I’ll note again that since Seedream cannot fix the seed value, precise controlled experiments are impossible🧪)

【Experiment Details】

① Generate with the prompt: "Anime style, watercolor painting finish, girl, blue eyes with light particles, black hair glowing light blue, blue and black butterfly-shaped hair ornament, blue and black gothic lolita-style costume, grassland, white and pink cosmos flowers in the foreground, glowing autumn leaves trees in the background, emerald and light blue gradient sky, sparkly, lens flare, rim light."

② Generate with the prompt: "Anime style, watercolor painting finish, girl, blue eyes with light particles, black hair glowing light blue, blue and black butterfly-shaped hair ornament, blue ribbon tie, lace, frills, blue and black gothic lolita-style costume, grassland, white and pink cosmos flowers in the foreground, glowing autumn leaves trees in the background, emerald and light blue gradient sky, sparkly, lens flare, rim light."

③ Generate with the prompt: "Anime style, watercolor painting finish, girl, blue eyes with light particles, black hair glowing light blue, blue and black butterfly-shaped hair ornament, blue ribbon tie, blue and black gothic lolita-style costume with lace and frills on the collar, hem, and sleeves, black ribbon around the waist, grassland, white and pink cosmos flowers in the foreground, glowing autumn leaves trees in the background, emerald and light blue gradient sky, sparkly, lens flare, rim light."

※As before, images are compressed due to capacity constraints

Previous image (before improvement) ↓

The clothing looks a bit baggy😅

Method ① ↓

Compared to before improvement, it’s simpler and more coherent! Even just this, the details are well rendered🔍️

Method ② ↓

Compared to before improvement, the lace and frills around the collar were generated without errors! Also, the shape of the ribbon tie is neater than before🎀

Method ③ ↓

Although there are some oddities on the front of the skirt, the generation errors are less noticeable than before improvement, and the overall silhouette is more refined🌟

In other words, all three were successes!

【In Conclusion】

Based on the results above, to generate clothing without errors,

① First, try generating with a simple prompt,
② If you want to add elements, detail them in bullet points in the prompt,
③ Also add elements to stabilize the composition and silhouette.

This seems to increase the chances of success!

Also, the suggestion from 日本及大和號戰艦的应援会™ that I couldn’t try this time—"Change the color combinations, light source positions, camera angles, etc., one by one for comparison"—sounds interesting and would be useful beyond Seedream!
(Considering cost, it seems safe to first try with a free model☺️)

By the way, which image do you like best? Personally, I prefer image ② with the cute smile and image ③ with the luxurious feel😆