Generation

Prompt Writing Guide

The quality of your prompt directly affects the quality of the generated content. Here's how to write prompts that get the best results.

How Prompting Works in Tesni

Unlike traditional AI tools, you don't write one big prompt. Tesni assembles the final prompt automatically from your connected nodes - Character description, Scene settings, Product details, and any additional text. You just fill out each node naturally.

Character Descriptions

Be specific about physical appearance:

Age & ethnicity - "25-year-old East Asian woman"
Hair - "shoulder-length dark brown hair with subtle highlights"
Build - "athletic build", "slim", "average"
Clothing - "wearing a white linen shirt and blue jeans"

Scene Settings

Use the Scene node dropdowns for lighting, camera angle, and mood - or add custom text for the environment:

Environment - "modern kitchen with marble countertops", "outdoor café in Paris"
Action - "pouring coffee", "holding the product with both hands"

Examples

✓ Good Character + Scene

Character: "28-year-old woman, warm brown skin, curly black hair, wearing a cream knit sweater."
Scene: "Cosy living room, golden hour light through sheer curtains, holding a ceramic mug of tea, smiling softly."

✗ Vague

"A person in a room"

Model-Specific Tips

FLUX Kontext Character - Best for consistent identity. Upload a clear, front-facing face photo.
FLUX Kontext Multi - Use when combining character + product reference images.
Recraft V4 - Great for product flat-lays and scenes with text overlays.
FLUX Schnell - Best for quick drafts. Keep prompts simple, iterate fast.

Image Order Matters (All Multi-Image Models)

When using any model that accepts multiple images — FLUX Kontext Multi, Nano Banana 2 Edit, and Seedream V5 Lite Edit — the order you connect your images determines how they are referenced. This is critical for getting the result you want.

How Image Slots Work

Image 1 — The primary subject. Typically your character reference (face, body, identity to preserve).
Image 2 — The secondary element. Typically the clothing, product, or prop you want applied to the character.
Image 3 (if supported) — An additional element such as a background scene or extra accessory.

✓ Correct Order

Image 1: Character photo → Image 2: Clothing flat-lay
Prompt: "The woman from image 1 wearing the black dress from image 2..."

✗ Wrong Order

Image 1: Clothing flat-lay → Image 2: Character photo
Result: The model treats the clothing as the primary subject and may distort the character or ignore identity.

Rule of thumb: Across all multi-image models, the first connected image is always the primary subject or base, and the second is the element being applied. For Kontext Multi, reference them explicitly using "image 1" and "image 2" in your prompt. For edit models like Nano Banana 2 and Seedream V5, the first image is the base to edit and the second is the design reference.

Preserving Body Proportions

Multi-image models can sometimes "normalise" body proportions when compositing a character into new clothing. To preserve specific features:

Repeat the constraint — Mention the feature you want preserved at the start, middle, and end of your prompt.
Use negative instructions — "Do NOT reduce, shrink, or alter..." reinforces the constraint.
Add physical descriptors — Words like "straining," "tight," "pulled taut" give the model concrete visual targets.
Use tighter framing — "Upper body and hips" gives more pixel budget to key areas vs. full body shots.
End with a CRITICAL section — Place your most important constraint at the very end of the prompt. Kontext models weight the end of the prompt heavily.

Was this helpful?

Pipelines Keyboard Shortcuts

Model-Specific Tips

FLUX Kontext Character - Best for consistent identity. Upload a clear, front-facing face photo.

FLUX Kontext Multi - Use when combining character + product reference images.

Recraft V4 - Great for product flat-lays and scenes with text overlays.

FLUX Schnell - Best for quick drafts. Keep prompts simple, iterate fast.

Image Order Matters (All Multi-Image Models)

How Image Slots Work

Image 1 — The primary subject. Typically your character reference (face, body, identity to preserve).
Image 2 — The secondary element. Typically the clothing, product, or prop you want applied to the character.
Image 3 (if supported) — An additional element such as a background scene or extra accessory.

✓ Correct Order

Image 1: Character photo → Image 2: Clothing flat-lay
Prompt: "The woman from image 1 wearing the black dress from image 2..."

✗ Wrong Order

Image 1: Clothing flat-lay → Image 2: Character photo
Result: The model treats the clothing as the primary subject and may distort the character or ignore identity.

Preserving Body Proportions

Multi-image models can sometimes "normalise" body proportions when compositing a character into new clothing. To preserve specific features:

Repeat the constraint — Mention the feature you want preserved at the start, middle, and end of your prompt.

Use negative instructions — "Do NOT reduce, shrink, or alter..." reinforces the constraint.

Add physical descriptors — Words like "straining," "tight," "pulled taut" give the model concrete visual targets.

Use tighter framing — "Upper body and hips" gives more pixel budget to key areas vs. full body shots.

End with a CRITICAL section — Place your most important constraint at the very end of the prompt. Kontext models weight the end of the prompt heavily.