3 Ways to Improve Seedream's Anime-Style Image Generation Issues

My 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 details remained.

So, I received various advice in the comments, and based on that, I’m taking another shot at creating 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 how the prompt was written, and I narrowed my focus to improving that.

Based on the suggestion from 日本及大和號戰艦的应援会™:
“For the costume, first generate it with a simple design, then gradually add details like lace and frills to compare,”
and the observation from こばと🕊️_ダヴ(DOVE):
“Writing the prompt in bullet points makes it 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) ⬆️

Method ③ is an application of the tip I previously wrote for using Image V2: “Add a prominent item like a hat to stabilize the composition” ⚖️👒
(Refer to .)

This time, I’m experimenting by modifying the part of the previous prompt: “a blue ribbon tie, a blue and black gothic lolita-style outfit with lace and frills on the collar, hem, and sleeves.”

(As a note, Seedream cannot fix the seed value, so precise controlled experiments are impossible, but I’ll reiterate that 🧪)

【Experiment Details】

① Generate with the prompt: “anime style, watercolor painting style, girl, blue eyes with light particles, black hair glowing light blue, blue and black butterfly-shaped hair ornament, blue and black gothic lolita-style outfit, 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 style, 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 outfit, 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 style, 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 outfit 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, these are compressed due to capacity limitations

Previous image (before improvement) ↓

The clothing looks a bit baggy 😅

Method ① ↓

Compared to before improvement, it’s become 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 are generated without flaws! Also, the ribbon tie shape is neater than before 🎀

Method ③ ↓

Although there’s a slight unnaturalness on the front of the skirt, it’s not as noticeable as the poor generation before improvement, and the overall silhouette is more refined 🌟

In short, all three were successful!

【Finally】

From these results, to generate clothing without flaws:

① First, try generating with a simple prompt,
② If you want to include more elements, write the prompt in detailed bullet points, or
③ Add elements to stabilize the composition and silhouette.

This shows that following these steps likely increases the chances of success!

Also, the suggestion from 日本及大和號戰艦的应援会™: “Change and compare colors, light source positions, camera angles, etc., one by one,” which I couldn’t try this time, sounds interesting and seems useful beyond just Seedream!
(Considering costs, it might be safer to first test with free models ☺️)

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