Telestyle V2 Style Transfer: The Essentials

Last updated: July 21, 2026

Telestyle V2 Style Transfer by Tele-AI borrows the look of any reference image, its artistic style, materials, lighting, and color palette, and applies it onto your own image while keeping the original composition fully intact. Version 2 removes a key V1 limitation and handles any pairing, from photo to photo and photo to painting through sketch to realism.

What It Does

You give the model two images. The Content Image sets what the result shows: its subject, layout, and composition are preserved. The Style Image sets how the result looks: its style, materials, lighting, and colors are transferred onto the content. The model keeps the structure of your content locked and repaints it in the visual language of the reference.

An optional auto-description step reads both images and feeds written descriptions into the process, which helps the model hold on to the subject while capturing the reference style more faithfully. You can dial the effect up or down with a single Style Strength slider, choose from six output shapes up to 1024px, and use a built-in speed booster for faster results.

Parameters

Telestyle V2 Style Transfer

  • Content Image (required): the image you want to restyle. Its subject and composition are kept; only the look changes.

  • Style Image (required): the image whose look you want to borrow. Its style, materials, lighting, and colors are applied to your content.

  • Style Strength (0 to 4, default 1): how strongly the reference style is applied. Higher values push the styling harder; lower values keep more of the original look. The examples below use values around 1.5 to 1.8 for a pronounced effect.

  • Resolution (default 1:1 HD, 1024x1024): the size and shape of the result, with six options from square to portrait and landscape.

  • Describe Content and Describe Style (both on by default): automatically analyze each input and feed a written description into the process to improve subject and style fidelity.

  • Use Lightning (on by default) and Lightning Strength: a speed booster that generates results faster. Lightning Strength only matters when the booster is on.

  • Inference Steps (2 to 50, default 4): how many refinement passes the model makes. More steps can improve quality but take longer.

  • Guidance (0 to 20, default 1): how closely the result follows the model's internal prompt.

  • Acceleration (None or Regular, default None): an optional speed boost. Regular is faster; None prioritizes quality.

  • Negative Prompt: things to keep out of the result, for example text, watermark, or blurry.

  • Images count (1 to 4) and Seed: how many results to create at once, and a number that makes a result repeatable when reused with the same settings.

Examples

In each row, the Content Image on the left keeps its composition, the Style Reference in the middle supplies the look, and the Result on the right combines the two.

Content Image

Style Reference

Result

A photo of a knight on horseback repainted in the impressionist palette and brushwork of a water-lily painting (Style Strength 1.5) · Open on Scenario

Content Image

Style Reference

Result

A rendered living room restyled with the ornate colors and floral motifs of a decorative painting (Style Strength 1.6) · Open on Scenario

Content Image

Style Reference

Result

A hummingbird photo restyled into a bold, high-saturation color palette while its pose and composition stay intact (Style Strength 1.6) · Open on Scenario

Tips for Better Results

  • Choose a Style Image with a strong, consistent look. The clearer the reference palette, materials, and brushwork, the more recognizable the transfer.

  • Use Style Strength to tune the effect. Start near the default and raise it toward 1.5 to 1.8 for a bolder restyle; lower it if the subject starts to dissolve.

  • Keep Describe Content and Describe Style on. The auto-description step helps preserve your subject and capture the reference style more faithfully.

  • Pick a Content Image with a clear, readable composition, since that structure is what the model preserves.

  • For maximum quality, turn the Lightning booster off or raise Inference Steps; for speed, keep Lightning on or set Acceleration to Regular.

  • Add a Negative Prompt such as text, watermark, blurry to keep unwanted artifacts out of the result.

  • Set a Seed when you find a look you like, so you can reproduce it or iterate on nearby variations.

Known Limitations

  • Composition is preserved, but fine details of the subject can shift as the image is repainted in the reference style.

  • Very high Style Strength can overpower the subject and wash out important detail. Back off if the content becomes hard to read.

  • Output resolution tops out at 1024px, so it is not intended for large-format final renders without a later upscale.

  • This is a global restyle, not a precise local editor. It changes the overall look rather than editing a specific region.

  • Results vary between runs unless you fix a Seed and keep the other settings the same.

Use Cases

  • Concept art exploration: try many art directions on the same composition in minutes.

  • Mood boards: unify a set of references under one consistent visual style.

  • Editorial restyling: give an existing image a new palette, lighting, or medium.

  • Look development: pitch and compare visual directions before committing to a final pipeline.

  • Consistent art direction: apply the same reference style across a batch of assets.