An experienced prompt engineer (promptologist) is required, well-versed in Google's graphic and multimodal models (Gemini, Nano Banana / Imagen 3) Key objective: Develop two universal, template-based prompts The prompts must work with any incoming photographs without the need to manually rewrite the text for each new case. The generation result must be crystal realistic, without visual artifacts of AI (“glossy” plastic skin, unnatural shine, blurred details). The photographs should look like frames from real life. Task essence in detail: The project consists of two phases, each requiring a ready universal prompt (with variables for API/n8n):Phase 1: Generation of variable “types” (70–80% similarity) Input: One original photo of a real person (any gender, nationality, age). Task: A universal prompt that takes the original photo and generates 10 new, different personalities of the same type. Requirements: Similarity 70–80%: The model should not completely copy the person but must strictly maintain race, nationality, eye color, skull shape, and distinctive features, position in the photo (if it’s just a person’s face, the generated photos should also be). Controllability: The prompt should allow adjusting the degree of similarity/variability (for example, by changing parameters or passing a variable value of similarity). Realism: Microtexture of skin, pores, natural irregularities, realistic lighting.Phase 2: Universal Inpainting / Face Swap Input: Original photo scene New face generated in Phase 1. Task: A universal prompt for replacing the face in the photo scene with the new face while preserving the environment. Requirements: Complete universality: The prompt must work correctly on any types of scenes (different lighting, angles, quality of the original) without manual adjustment of the text for each frame. Scene preservation: Changing the background, clothing, lighting, pose, or hairstyle of the original scene is prohibited. Only geometry and facial features are replaced. Seamlessness: The new face must automatically adjust to the lighting, shadow, and color of the target frame. No seams, blurs, or “mask” effects. The prompts must be formatted as dynamic templates with variables (parameters) that can be easily substituted via JSON/n8n HTTP Request node or standard integrations. The freelancer must provide detailed instructions on calling: which specific API parameters (temperature, top_p, guidance scale, style weight, etc.) ensure stable results during automatic runs. Requirements for the performer Deep experience with Gemini / Nano Banana API: Understanding how the model accepts multimodal references (Image-to-Image / Inpainting) via API. Portfolio: Availability of real examples of photorealistic generation of people. (Anime, 3D, and “glossy” AI portraits are not considered). Understanding of automation: Experience in composing prompts for pipelines (n8n, Make, Python scripts), where stability and universality of response are important. When responding, you must be ready for a test process, meaning you generate 2-3 similar faces, then perform a test face swap. If you can write stable universal prompts without AI artifacts — we look forward to your response! Start your message with the word “Banana” when responding
100 USD
16 proposals
3 August