Watercolor Portrait From 3D Character Reference Prompt

Here's the thing.

@Soaima_Ai labels this one Seedance 2.0, then hands over the prompt. The label matters because of what the post is quoting. The quoted tweet above it is the upstream step, a GPT image 2 prompt that generates a photorealistic 3D character portrait of a young woman beside a cute oversized blue fluffy bird. This post takes that kind of render as the uploaded reference and repainted it, same girl, new medium.

The result is the attached image. A watercolor fashion portrait of the same young woman, three-quarter over-the-shoulder pose, looking back toward the camera with a warm confident expression. Long flowing dark brown-black hair with subtle deep blue and warm orange painted highlights, expressive hazel-brown eyes, rosy cheeks, elegant dangling pearl earrings. Loose white outfit with artistic orange, yellow, blue and coral watercolor brushstrokes.

The style brief is precise about the medium. A sophisticated hand-painted watercolor and ink illustration, visible brush textures, delicate paint splashes, flowing strokes, artistic color bleeding around the edges. The background stays mostly white, with abstract watercolor splashes in sky blue, orange, beige, navy blue and soft pink drifting around the figure.

The part most style-transfer prompts skip is the negative prompt, and this one ships a full block. No blurry face, distorted eyes, extra fingers, deformed hands, bad anatomy, plastic skin, 3D render, cartoonish face, oversaturated colors, low resolution, duplicate features, text, watermark, logo or screenshot UI. Note the second item in that list, no 3D render. The input is a 3D render, the whole point is that the output must stop being one.

The prompt, as posted.

Create a beautiful high-quality watercolor portrait of a young woman, based on the uploaded reference image. She has long, flowing dark brown/black hair with subtle deep blue and warm orange painted highlights, soft natural facial features, expressive hazel-brown eyes, neatly shaped eyebrows, a gentle smile, rosy cheeks, and elegant dangling pearl earrings.

She is shown in a three-quarter over-the-shoulder pose, looking back toward the camera with a warm, confident expression. She is wearing a loose white outfit with artistic orange, yellow, blue, and coral watercolor brushstrokes.

The entire artwork should have a sophisticated hand-painted watercolor and ink illustration style, with visible brush textures, delicate paint splashes, flowing strokes, and artistic color bleeding around the edges. Background should be mostly white with abstract watercolor splashes in sky blue, orange, beige, navy blue, and soft pink.

Highly detailed face, beautiful eyes, realistic proportions, elegant composition, soft natural lighting, fine watercolor paper texture, artistic fashion portrait, premium digital watercolor painting, vibrant but harmonious colors, 4K, vertical portrait, 2:3 aspect ratio.

Negative prompt: blurry face, distorted eyes, extra fingers, deformed hands, bad anatomy, plastic skin, 3D render, cartoonish face, oversaturated colors, low resolution, duplicate features, text, watermark, logo, screenshot UI.

A small note on the original. The author is Soaima, posting as Soaima_Ai, the same account behind the snow-globe McDonald's clip, the Vaseline commercial and the watercolor autumn prompts in this library. The post is in English, the prompt ships in full in the tweet body, and the attached image is the watercolor output. The quoted tweet carries the upstream prompt in full, a GPT image 2 render of the same woman standing beside a large blue fluffy bird on a solid sky-blue studio background, so the two posts form a two-step pipeline. Generate the character render first, then run this prompt on it to pull the character out of the 3D world and into watercolor and ink. The medium-swap template generalizes, keep the pose and identity clauses and replace the watercolor style block with any other painting medium.

At the time of writing, the post sits at 23 likes, 5 retweets, 9 bookmarks and 822 views, with 21 replies under it. Original thread here, x.com/Soaima_Ai/status/2095811218523275552