Skip to main content

Style Generator

Create a full-screen style effect from a text prompt or reference images with Style Generator in Lens Studio AI, train it, and import it into your Lens.

  • Snapchat: fully supported
  • SPECS: partially supported
  • Camera Kit Web: not supported
  • Camera Kit Mobile: not supported
  • SnapchatFully supported
  • SPECSPartially supported
  • Camera Kit WebNot supported
  • Camera Kit MobileNot supported

Platform details

  • Snapchat — Fully supported.

  • SPECS — Partially supported.

  • Camera Kit Web — Not supported.

  • Camera Kit Mobile — Not supported.

Style Generator creates full-screen stylization effects for your Lens. It works across the entire frame, transforming people, objects, pets, and backgrounds into a new style.

Style Generator gives you full control over how the world looks through your Lens, with high-quality, varied results. Create anything from realistic looks, with refined textures, colors, and lighting, to bold artistic transformations, including 2D and 3D graphics, vector styles, anime, pop art, and more.

Open Style Generator in Lens Studio AI

  1. In the Lens Studio toolbar, select Lens Studio AI.

    The Lens Studio toolbar with the Lens Studio AI button outlined in red

    The Lens Studio AI panel opens on the GenAI Suite home, a grid of tool cards under Generate ML models and Assets.

  2. Select the Style Generator card, near the top of the grid beside AI Photo Duo. To see all the tools, scroll the panel.

    The Lens Studio AI panel on the GenAI Suite home, with Face Generator, Style Generator, and AI Photo Duo cards
    The end of the GenAI Suite tool grid, with cards including Selfie Attachments and Face Mask

Style Generator is also an Asset Library plugin. If the Style Generator card isn't in the grid, install the Style Generator plugin from the Asset Library. Once installed, you can also open it from the Asset Browser.

Find your way around the Style Generator tool

The Style Generator tool has two areas:

  • Creation Panel: the tools for creating and editing effects.

    The Style Generator Creation Panel outlined beside the Gallery
  • Gallery: every effect you created, with its status.

    The Style Generator Gallery listing created effects with their statuses

Choose a Style Generator effect type

Style Generator has three effect types, under Effect type in the Creation Panel:

  • Enhanced suits a wide range of cartoon and artistic styles while preserving the person's identity. It has more customization options, for immersive results. Depending on the settings and prompts, the output ranges from realistic to highly stylized: from lifelike renderings to bold graphic illustrations and cartoons.

  • Cartoonish turns the user into a character that looks like it came from an animated movie, a comic book, or a piece of art. For examples of effects this type creates, see Cartoonish effect examples.

  • Realistic applies transformations while keeping a natural, lifelike appearance.

    EnhancedCartoonishRealistic

Create an effect with Style Generator

The creation process is the same for every effect type. The main difference is that Enhanced has more settings and accepts an Effect Prompt and an Image Reference together, while Cartoonish, for example, accepts only one input at a time: either a text prompt or an image prompt.

You can generate up to 5 effects at once ("A maximum of 5 effects can be generated at once") and train up to 5 models at once ("A maximum of 5 models can be trained at once").

The following steps use the Enhanced effect type.

  1. To try one of the default prompts and get familiar with the model, select Surprise me.

    The Surprise me link above the Effect Prompt field, with a default prompt filled in
  2. To use your own prompt, describe the effect you want in the Effect Prompt field, up to 500 characters.

    To get the best results, see Write prompts for Style Generator.

    A custom description in the Effect Prompt field, with its character count out of 500
  3. To guide the generation closer to your vision, add up to 5 images in Image Reference.

    You can use an Effect Prompt, an Image Reference, or both, whichever works best for your case. To choose good reference images, see Choose reference images for Style Generator.

    An image added to the Image Reference field of the Creation Panel
  4. Adjust the settings. They can change the result significantly and help you match your vision:

    • Reference Strength: how strongly the effect follows the Image Reference. A higher value brings the result closer to the Image Reference, but makes it less similar to the original user photo.
    • Seed: controls randomness. Use the same number to recreate the same look, or try different numbers for new variations.

    To learn more, see Tune Style Generator settings.

    The Reference Strength slider and the Seed field in the Creation Panel
  5. When your prompt is ready, select Generate previews.

    The Generate previews button at the bottom of the Creation Panel

    A new tile appears in the Gallery, with a loading indicator in its corner that shows the progress. Depending on the effect type, Lens Studio estimates up to 5 minutes, or 10–15 minutes, to generate the previews. You can close the tool and come back later.

    A new effect tile in the Gallery with a progress indicator in its corner
  6. When the previews are ready, select the tile in the Gallery to open the details page. Use the arrows on the sides to preview the effect on different models. To see the original image the effect was applied to, select the button in the bottom-right corner.

    The effect details page with arrows on the sides and the original-image button in the bottom-right corner
  7. If the preview doesn't match your expectations, select Copy Settings to adjust your prompts or seed.

    The Copy Settings button on the effect details page

    Make as many tweaks as you need. Each change appears in the Gallery as a new effect and leaves the current one unchanged, so you can always go back and compare the updated effect with the original.

  8. When you're happy with the preview, select Train model. You can track the training status in the Gallery. Depending on the effect type, Lens Studio estimates 3–5 hours, or up to 2 hours, to train the model. You can close the tool and come back later.

    The Train model button on the effect details page
  9. When training is complete, select Import on the effect's tile in the Gallery, or select Import to Project on the details page, to add the effect to your project.

    The Import button on a trained effect's Gallery tile and the Import to Project button on the details page

Preview and test a Style Generator effect in your Lens

When you import an effect, Style Generator builds a full-screen camera effect in your scene: an orthographic camera, a Device Camera Texture, a Full Frame Region, and an ML component that applies the trained model.

Check the result in the Preview panel and continue building your Lens.

The Style Generator Component in the Inspector, with Device Camera Texture as its Input Texture, and the effect applied in the Preview

When building for Snapchat, save your project and push the Lens to Snapchat to test it on a device. See Pairing to Snapchat.

When you're satisfied with the result, see Submitting Your Lens to learn how to share your creation.

To combine Style Generator with Animated Selfie Attachments, Head Generator, and the Zoom component, see Animated Selfie Attachments & More. To combine Style Generator with Face Generator, Face Animator, and Garment, see Style Gen with Face Gen & More.

Write prompts for Style Generator

For the best results, combine an Image Reference with an Effect Prompt.

Write an Effect Prompt for Style Generator

You can use short and long prompts to guide the generation. It usually works best to start with a short prompt and refine it as you go: add adjectives, specific details, or extra descriptions to bring the result closer to the look you want. Style Generator interprets your prompt literally.

  • Focus on the essentials. Describe only the changes you want to make, and leave out unnecessary or generic details.
  • Skip the background. Mention the background only when it's part of your design.
  • Use descriptive words. Adjectives like hand-drawn, 2D, 3D, cartoon, or realistic help you control the style of the generated image.
  • Write long prompts for distinctive designs. More detailed prompts help you create stronger, more distinctive designs. Combine them with fine-tuned settings for precise, unique results.
  • Separate phrases with commas. Divide long descriptions into short, context-specific phrases separated by commas: cute cartoon anime, kawaii romantic colors, immersive, photorealism.
  • Weight words. Special symbols in your Effect Prompt control how much influence specific words or phrases have on the result, emphasizing or reducing their impact. See Control word influence with prompt weighting.

Descriptors for Style Generator prompts

  • Style: the style you want to achieve.
    • Example: oil painting portrait, stoic, oil texture, pixel style, game consoles, graphic design, retrofuturistic, vintage style, vector art, detailed outlined minimalism, comic art, hand drawn illustration, poster, flat illustration, children's picture book illustration, 3D papercuts, surrealistic cartoons, inspired by paper cut-outs art and paper sculptures, punk illustration, nightcore, flat outlined illustration, dnd art style, digital illustration, CG, risography effect, charcoal courtroom sketch
  • Colors: the colors and palettes you want to emphasize.
    • Example: dark sepia color pallet, illustration in bright colors, style in full pastel colors, realistic usage of light and color, pastel color palette theme, nightcore, magical pink, soft colors
  • Patterns: pattern descriptors add a sense of texture and depth, making flat surfaces look more dynamic and interesting.
    • Example: heavy thick brush strokes, thatch textures, geometric patterns, floral patterns, herringbone patterns, chevron patterns, polka dots, stripes, plaid, houndstooth, animal prints, paisley, abstract patterns, crosshatch, swirls, damask, zigzag, checkerboard, gradient, lattice, argyle, marble, wave patterns, brick patterns, mesh, hexagonal patterns, brocade, tartan, quilted patterns, leaf patterns
  • Light, shading, and lines: descriptors for special lighting and shading details, when you need them.
    • Example: realistic usage of light, natural light, soft focus, blur, detailed shading, soft flush, flat shading, precise lines, with clear lines, anti-form, flat outlined illustration
  • Mood: the mood, when you need it.
    • Example: happy, cheerful, calm, relaxed, mysterious, intriguing, energetic, excited, sad, melancholic, romantic, dreamy, tense, suspenseful, playful, fun, warm, cozy, lonely, isolated, adventurous, bold, reflective, thoughtful, angry, intense, joyful, celebratory, peaceful, harmonious

Special words for Style Generator prompts

  • Close-up: frames the image tightly around the subject, highlighting specific features and details.
  • High quality: makes the generated image clear, sharp, and visually appealing.
  • Detailed: asks for many small, intricate elements.
  • Flat shading: asks for shading without gradients, for a more stylized, simpler look.
  • HD: combines high resolution and extreme detail for a highly realistic output.
  • Maximalistic: asks for a style rich in detail and complexity, often with elaborate, decorative elements.
  • High definition: creates the image in high resolution, for more clarity and detail.
  • Fine luster: adds a touch of shine or gloss, for a polished, refined look.

Choose reference images for Style Generator

Use high-quality reference images, ideally of a human or character with refined textures. The quality of the reference image, including its overall aesthetics and the subject it shows, directly affects how good the result looks. Combine the image reference with a prompt that names the elements you want to transfer from the image to the effect.

Examples of good image references:

A good image reference: a painted pine forest with a stream in bright colors
A good image reference: an anime-style character in a pastel city at night
A good image reference: iridescent, glittering fabric in pastel colors
A good image reference: a detailed pencil drawing of a face

Combine an Effect Prompt and an Image Reference in Style Generator

Combining an Image Reference with an Effect Prompt expands your design possibilities and lets you explore styles that differ from the original image. To recolor the outfit or add more details, enter a custom Effect Prompt.

Compare Style Generator results by prompt combination

The following examples show how the result changes with the combination of prompts. They use this Effect Prompt:

realistic, glitter, sparkles, golden hour lighting, nostalgic, pink magic glowing background
  • Effect Prompt only:

    Original user imageEffect PromptProcessed user image
    The original user photo before the effect
    The Creation Panel with only the Effect Prompt filled and no Image Reference
    The user photo processed with the Effect Prompt alone
  • Image Reference only:

    Original user imageImage ReferenceProcessed user image
    The original user photo before the effect
    The Creation Panel with only an Image Reference added and no Effect Prompt
    The user photo processed with the Image Reference alone
  • Effect Prompt + Image Reference:

    Original user imageEffect Prompt + Image ReferenceProcessed user image
    The original user photo before the effect
    The Creation Panel with both the Effect Prompt and an Image Reference filled
    The user photo processed with the Effect Prompt and Image Reference together

Tune Style Generator settings

Set the Reference Strength in Style Generator

Reference Strength controls how strongly the effect follows the Image Reference. A higher value brings the result closer to the Image Reference.

Image ReferenceRef. Strength = 4.0Ref. Strength = 7.0Ref. Strength = 9.0
The anime-style character image reference
Result with Reference Strength 4.0, loosely following the anime reference
Result with Reference Strength 7.0, closer to the anime reference
Result with Reference Strength 9.0, closest to the anime reference
Image ReferenceRef. Strength = 4.0Ref. Strength = 7.0Ref. Strength = 9.0
The pencil drawing image reference
Result with Reference Strength 4.0, loosely following the drawing reference
Result with Reference Strength 7.0, closer to the drawing reference
Result with Reference Strength 9.0, closest to the drawing reference
Image ReferenceRef. Strength = 4.0Ref. Strength = 7.0Ref. Strength = 9.0
The glittering fabric image reference
Result with Reference Strength 4.0, loosely following the glitter reference
Result with Reference Strength 7.0, closer to the glitter reference
Result with Reference Strength 9.0, closest to the glitter reference

Choose a Seed in Style Generator

The Seed controls randomness, and choosing the right seed is often crucial for the best results. Use the same number to recreate the same look, or try different numbers for new variations.

The following examples show how the result changes with the seed for this prompt combination.

Effect Prompt:

2d art, detailed illustration, cute, soft colors, blush, shadows, ginger, sunset

Image Reference:

The painted pine forest used as the Image Reference
Seed: 2570Seed: 32213Seed: 3904Seed: 40907
Result with Seed 2570
Result with Seed 32213, different from the other seeds
Result with Seed 3904, different from the other seeds
Result with Seed 40907, different from the other seeds

Cartoonish effect examples

These Cartoonish effects show what the effect type can create from a prompt.

Prompt: wooden dummy doll in a cardboard box cityPrompt: black and white chalk drawing
Cartoonish effect from the prompt wooden dummy doll in a cardboard box city
Cartoonish effect from the prompt black and white chalk drawing
Prompt: knotted handmade doll out of fabric and yarnPrompt: curly hair, volume, curls, waves, hairdo, tiny look
Cartoonish effect from the prompt knotted handmade doll out of fabric and yarn
Cartoonish effect from the prompt curly hair, volume, curls, waves, hairdo, tiny look
Prompt: fine art portrait from 18 century, museum masterpiece, floral background, pastel colorsPrompt: masterpiece, face portrait painting, impressionism, watercolor, in the style of Claude Monet
Cartoonish effect from the prompt fine art portrait from 18 century
Cartoonish effect from the prompt face portrait painting in the style of Claude Monet
Prompt: neon lights, long exposure of light, luminous, radiantPrompt: cinematic red light, bokeh, cartoon face, white particles in background
Cartoonish effect from the prompt neon lights, long exposure of light, luminous, radiant
Cartoonish effect from the prompt cinematic red light, bokeh, cartoon face

Control word influence with prompt weighting

Special symbols in a Style Generator Effect Prompt control how much influence specific words or phrases have on the result, emphasizing or reducing their impact.

Append +, -, or a weight between 0 and 2 (1 is the default) to a word or phrase to increase or decrease its attention during generation. Style Generator supports this syntax:

  • Single words (no parentheses): retro comic art+
  • Single or multiple words (with parentheses): retro comic (art)+, retro comic (art)-, retro (comic art)+, retro (comic art)-
  • Stronger emphasis with more symbols: retro comic (art)++
  • Nested weighting: retro (comic art+)++, where "art" effectively receives "+++"
  • Explicit numeric weights: retro (comic art)1.1, retro (comic (art)1.3)1.1

Weight equivalents:

  • + = 1.1
  • ++ = 1.1²
  • +++ = 1.1³
  • - = 0.9
  • -- = 0.9²

The following examples start with the prompt retro comic art and show how the result changes when you increase or decrease the model's attention to the word retro.

Using - to reduce how retro the result looks:

retro comic artretro- comic artretro-- comic artretro--- comic art
Result for retro comic art
Result for retro- comic art
Result for retro-- comic art
Result for retro--- comic art

Using + to increase how retro the result looks:

Use at most three plus signs: more than three can produce broken visuals.

retro+ comic artretro+++ comic artretro++++ comic artretro+++++ comic art
Result for retro+ comic art
Result for retro+++ comic art
Result for retro++++ comic art
Result for retro+++++ comic art

You can also change the balance between different parts of a prompt:

retro comic art, mountainsretro comic art, mountains+retro comic art, mountains++retro comic art, mountains+++
Result for retro comic art, mountains
Result for retro comic art, mountains+
Result for retro comic art, mountains++
Result for retro comic art, mountains+++, with mountains in the background and on the sweater

Or, alternatively, with more art:

retro comic art+, mountainsretro comic art++, mountainsretro comic art+++, mountainsretro comic art++++, mountains
Result for retro comic art+, mountains
Result for retro comic art++, mountains
Result for retro comic art+++, mountains
Result for retro comic art++++, mountains

Single or multiple words with parentheses:

retro comic (art)+retro comic (art)++retro (comic art)+retro (comic art)-
Result for retro comic (art)+
Result for retro comic (art)++
Result for retro (comic art)+
Result for retro (comic art)-

Nesting:

retro (comic art+)++(retro)+ comic art++(retro+ comic)+++ art
Result for retro (comic art+)++
Result for (retro)+ comic art++
Result for (retro+ comic)+++ art

With explicit numbers:

retro (comic art)1.1retro (comic (art)1.3)1.1
Result for retro (comic art)1.1
Result for retro (comic (art)1.3)1.1
retro (comic (art)1.3)1.1, mountains, (clouds)1.3(no explicit numbers): retro comic art, mountains, clouds
Result for retro (comic (art)1.3)1.1, mountains, (clouds)1.3
Result for retro comic art, mountains, clouds, with no weights