DreamShaper is one of the most popular versatile models that can generate almost any image. It excels in realistic artwork, but can also handle many other art styles. The model's ease of use and capabilities make it a great choice for any user.

SeaArt - Model Page

What This Model Excels At

• Crafting photorealistic/3D render style versatile images.
• Generating female/male human images.
• Generating animal/object/landscape type subjects.
• Crafting images with SFW/NSFW content.
• Crafting various styles. e.g. "illustration, abstract, sketching, painting etc."
• Crafting themes like fantasy, medieval, sci-fi etc.
• Great lighting/shadow usage.
• Great depth/focus usage.
• Recognizes most of art/artist styles.

Cautions For Optimal Usage

• Recommended to use "BadDream, UnrealisticDream" embeddings in your negative prompt. User can choose to not use them but they increase quality to highest levels. "BadDream" is a casual negative embedding while "UnrealisticDream" is focus on improving realism.
• `anime coloring, anime screencap` can be used in front of prompt to craft anime style images, unfortunately model struggle to maintain anime style and towards to generate 3D render images sometimes.
• Distance reduces quality of subject, prefers "close-up, portrait, medium range" shots.
• Better with single subject images.
• Using lighting, shadow and focus terms increase realism and quality of images.
• Work great with additional quality modifiers like `8k, UHD, trending on deviantart` etc.
• Works pretty nice with many various style/character LoRA.
• Versions have certain differences, this guide focus on latest version (v8) but it will include noticeable differences between each version.
• Recommended to not use "Restore faces". ADetailer is great option for improving image's detail and quality.

Recommended Settings

• Steps: 20-30
• CFG Scale: 7-9 (can be increased to 12 but not necessary for most situations)
• Sampler: DPM++ SDE Karras, DPM++ 2M Karras
• Clip Skip: 2
• VAE: User's choice (depending on creation and personal preference, model works great with all of them)

Model doesn't require "Advanced Repair" generally, using is up to your decision.

• Upscale: R-ESRGAN 4x+, 4x-UltraSharp, 8x_NMKD-Superscale_150000_G
• Upscale Sampling Times: 10-20
• Redraw Noise Intensity: 0.3-0.6

Suggested Resources

LoRA

- Add More Details - Detail Enhancer / Tweaker (细节调整) LoRA

Embedding

- Fast Negative Embedding V2

- BadDream + UnrealisticDream (Negative Embeddings)

Trigger words of embeddings must be used on negative prompt.

Prompt Style

• Use tag based prompts instead of natural language prompts.
• Can work with both of short and long prompts, usually long prompts have better results compared to short ones.
• Use descriptive prompts to achieve desired results.
• Can understand many of various terms from photography to drawing technique.
• Quality Modifiers have great impact to quality and details of output.
• Using references (artist, website etc.) has noticeable improves on results, recommended to use.
• Great usage of `depth of field, bokeh, sharp focus` like terms.
• Use `anime coloring, anime screencap` in front of prompt to make anime renders.

Recommended Prompt Order

1. Base Composition or Subject (e.g. `ultra realistic portrait of a woman`-`3D rendered car` etc.)
2. Angle/View
3. Medium/Style
4. Art/Artist References
5. Quality Modifiers
6. Details of Composition
7. Lighting/Shadow
8. Focus/Depth
9. Additional Modifiers

Prompt Examples

Fungal Wastes

Positive Prompt

`8K render of a mysterious forest, abstract, surreal, (highly intricate:1.2), (masterpiece, best quality:1.3), (ultra detailed:1.4), UHD, 32K, (amazing landscape:1.3), (ethereal atmosphere:1.2), (color blending), glow effects, (glowing colors:1.2), {purple theme|(blue theme:1.2)|black theme}, (scattering:1.2), ((glowing flowers, glowing mushrooms)), (shimmer particles:1.3), (glowing, shimmer:1.4), diffused lighting, (water reflection), (film grain:1.2), (depth of field:1.3), [dark ambient], (dream-like art, fantasy), ray tracing, (unreal engine, octane render:1.3), (VFX, volumetrics dtx)`

Negative Prompt

`BadDream, (FastNegativeV2:1.3), monochrome, faded`

Bloodlust

Positive Prompt

`portrait of a woman, vampire queen, illustration, sfw, fantasy, {(by Charlie Bowater)|(by Jesper Ejsing:1.2)|by Sam Bosma}, (masterpiece, best quality:1.3), (highly detailed:1.5), UHD, 8K, beautiful, pale skin, white skin, jawline, cheekbones, (red lips, red eyes), black hair, short hair, collarbone, leather cloak, black cloak, hooded, looking at viewer, snowy fields, moon, (amazing background), moonlight, backlighting, diffused light, soft focus, (depth of field), trending on artstation`

Negative Prompt

`BadDream, (FastNegativeV2:1.3), ugly, monochrome, (worst quality, worst image:1.5), (low detail:1.5), lowres, nsfw, (deformed eyes:1.3)`

Version Differences

• V4 is best at anime, can recognize "Danbooru" tags.
• V5 is best at photorealism, has noise offset.
• V6 has better LoRA support and more general style.
• V7 improves LoRA support, realism and NSFW.
• V8 is improved version of V7, better LoRA support, style control etc.

In general, V8 is recommended.

You can join SeaArt Discord Community for more content, also you will have opportunity to choose next model guides with your vote. You can use comments or SeaArt server #guides-feedback to leave feedback or suggestion. We look forward to seeing you on the server.

Happy Generations To Everyone!