xbrush.ai offers dozens of AI generation models across image, video, and audio — everything from completely free options to high-credit premium tools. The range is wide.
Every time someone asks "which model should I pick?", I feel the need for a clear reference. This guide is exactly that: a practical reference covering pricing, billing logic, style characteristics, and recommended use cases — all in one place.
Pricing basis: creditConfig v24 (2026-06-16) / Unit: credits / Active models only
Image Generation Model Comparison
Basis: 4 images, 1024×1024, UI-displayed credit cost
Free Models
| Model | Characteristics | Limit |
|---|---|---|
| Z-Image Free | Fast generation, lightweight and practical style | 10/day |
| Hunyuan 3.0 Free | Tencent Hunyuan base, strong East Asian aesthetic | 10/day |
Even free models can match paid ones for certain styles. The daily 10-use limit is fine for light testing or low-volume work.
Economy Models (10–50 credits / 4 images)
| Model | Credits (4 img) | Billing | Best For |
|---|---|---|---|
| SDXL 1.0 | 10 | perMegapixel | Quick prototypes, general use |
| Z-Image Turbo | 12 | perMegapixel | Cost-effective illustration, characters |
| Qwen Image | 16 | perImage | East Asian style — use English prompts |
| FLUX.1 DS | 20 | perMegapixel | Fine texture, photographic style |
| Qwen Image RE | 26 | perMegapixel | Qwen improved, stronger realism |
| FLUX.1 D | 42 | perMegapixel | High-quality photo realism |
The FLUX family excels at realistic styles. The Dev (D) variant outperforms Schnell (DS) in quality but costs more than double. The Qwen family performs reasonably well on East Asian prompts but does not fully understand Korean.
Balanced Models (47–100 credits / 4 images)
| Model | Credits (4 img) | Billing | Best For |
|---|---|---|---|
| Seedream 4.0 | 47 | byResolution (2K) | Illustration, webtoon feel |
| Wan 2.7 | 47 | perImage | Versatile, balanced output |
| Gemini 2.5 Flash | 48 | perImage | Text rendering, multilingual |
| Flux 2 Pro | 50 | perMegapixel | FLUX premium, commercial quality |
| Nano Banana ★ | 62 | perImage | Natural portraits, warm palette — full Korean support |
| Seedream 4.5 | 63 | byResolution (2K) | Seedream improved |
| GPT Image 2 ★ ⚠️ | 83 (medium) | byResolutionAndQuality | Precise instruction following, text — full Korean support |
★ Full Korean language support | ⚠️ GPT Image 2 quality tiers (4 img, 1K): low 10 / medium 83 / high 330 / 2K high 659 credits — verify your quality setting before generating.
Premium Models (148+ credits / 4 images)
| Model | Credits (4 img) | Billing | Best For |
|---|---|---|---|
| Hunyuan 3.0 Instruct | 148 | perMegapixel | High instruction accuracy, commercial ads |
| Nano Banana 2 ★ | 188 | byResolution (2K) | High-res portraits/product shots — full Korean support |
| Nano Banana Pro ★ | 234 | byResolution (2K) | Top-tier quality, advertising visuals — full Korean support |
| GPT Image 2 (high) ★ | 330 | byResolutionAndQuality | Precise instruction, text-heavy images — full Korean support |
★ Full Korean language support
Korean Language Support
Most image generation models on xbrush do not fully understand Korean. Even when you type in Korean, the model may translate it internally or lose linguistic nuance in the process.
Models with full Korean language understanding:
| Model | Category | Notes |
|---|---|---|
| GPT Image 2 | Image gen & edit | Strong at both text rendering and Korean instruction following |
| Nano Banana | Image gen & edit | Google-based, natural portraits and warm color palette |
| Nano Banana 2 | Image gen & edit | Best for high-resolution portrait photography |
| Nano Banana Pro | Image gen & edit | Top-tier quality for advertising visuals |
| Seedance 2.0 Fast | Video gen | 720p, accurately reflects Korean scene descriptions |
| Seedance 2.0 | Video gen | High-quality 720p, handles Korean story prompts |
All other models — FLUX, SDXL, Seedream, Wan, Hunyuan, Qwen, etc. — do not fully understand Korean. Results may diverge significantly from what a Korean prompt describes. English prompts produce more predictable results with these models.
The comparison below shows the same Korean-language prompt run through Qwen Image and FLUX 2 Pro:
Neither model fully processes the Korean input, and both show a meaningful gap versus English-prompt results.
Image Editing Model Comparison
Edit tab basis: 4 images, 1024×1024 — same billing logic as image generation
| Model | Credits (4 img) | Key characteristics |
|---|---|---|
| Hunyuan 3.0 Free Edit | FREE | Free inpainting, 10/day limit |
| Qwen Image Edit RE | 30 | Partial edits, cost-effective |
| Seedream 4.0 Edit | 47 | Illustration-style editing |
| Flux 2 Pro Edit | 50 | High-quality photorealistic editing |
| Gemini 2.5 Flash Edit | 48 | Edit images with embedded text |
| Nano Banana Edit ★ | 62 | Portrait editing, natural retouching — full Korean support |
| Seedream 4.5 Edit | 63 | Seedream improved editing |
| GPT Image 2 Edit ★ ⚠️ | 83 (medium) | Precise instruction editing, strong text — full Korean support |
| Hunyuan 3.0 Instruct Edit | 148 | Handles complex edit instructions |
| Nano Banana 2 Edit ★ | 188 | High-res partial editing — full Korean support |
| Nano Banana Pro Edit ★ | 234 | Top-tier editing quality — full Korean support |
★ Full Korean language support | ⚠️ GPT Image 2 Edit: medium 83 / high 330 credits
Video Generation Model Comparison (i2v)
Basis: default duration & resolution / noAudio pricing
Free Models
| Model | Default Duration | Characteristics |
|---|---|---|
| LTX 2.3 Free | 5 sec | Fast generation, good for testing and experimentation |
Economy Models (106–164 credits / clip)
| Model | Credits | Dur | Resolution | Per-sec | Characteristics |
|---|---|---|---|---|---|
| Hailuo 02 Standard | 106 | 6 sec | 768p | 17.6/s | Smooth motion |
| Kling v2.1 Standard | 110 | 5 sec | — | 21.9/s | Stable, versatile |
| Kling v2.5 Turbo Pro | 137 | 5 sec | — | 27.3/s | Kling cost-effective |
| LTX 2.3 | 141 | 6 sec | 1080p | 23.4/s | HD, stable |
| Wan 2.5 Preview | 150 | 5 sec | — | 30/s | Multi-purpose |
| Wan v2.2 14B | 153 | 5 sec | 720p | 30.6/s | Large model, balanced |
| Kling v3 Standard | 164 | 5 sec | — | 32.8/s | Kling v3 baseline |
Balanced Models (176–295 credits / clip)
| Model | Credits | Dur | Resolution | Per-sec | Characteristics |
|---|---|---|---|---|---|
| Kling v2.1 Pro | 176 | 5 sec | — | 35.1/s | Natural human motion |
| Kling v1.6 | 186 | 5 sec | — | 37.1/s | Proven Kling legacy |
| Hailuo 02 Pro | 188 | 6 sec | 1080p | 31.2/s | HD professional grade |
| Kling v3 Pro | 219 | 5 sec | — | 43.7/s | Latest Kling pro |
| Seedance 2.0 Fast ★ | 236 | 5 sec | 720p | 47.2/s | Fast high-quality — full Korean support |
| Wan 2.7 Video | 293 | 5 sec | 1080p | 58.5/s | HD balanced |
| Wan v2.5 Preview | 293 | 5 sec | 1080p | 58.5/s | Improved Wan |
| Seedance 2.0 ★ | 295 | 5 sec | 720p | 58.9/s | High-quality standard — full Korean support |
★ Full Korean language support
Premium Models (312+ credits / clip)
| Model | Credits | Dur | With audio | Characteristics |
|---|---|---|---|---|
| Veo3 Fast | 312 | 8 sec | 624 | Google Veo3, fast variant |
| Veo3.1 Fast | 312 | 8 sec | 624 | Veo3 improved fast variant |
| Kling v2.1 Master | 546 | 5 sec | — | Top-tier quality, cinematic |
| Veo3 | 624 | 8 sec | 1,248 | Google premium, 8 sec |
| Veo3.1 | 624 | 8 sec | 1,248 | Latest Veo3 iteration |
Kling v3 Pro with audio: 328 credits (5 sec) / Veo3 series: audio doubles the price
Audio Model Comparison
Music Generation
| Model | Credits / track | Characteristics |
|---|---|---|
| Default | 10 | Basic background music |
| Lyria 3 | 15.6 | Google Lyria, natural musicality |
| Lyria 3 Pro | 31.2 | Lyria high-quality, broad genres |
| Lyria 2 | 39 | Previous-gen Lyria, strong for specific styles |
Lipsync — 30-second basis
| Model | Credits (30 sec) | Characteristics |
|---|---|---|
| Default | 300+ | Basic lipsync |
| PixVerse Lipsync | 468+ | Natural mouth movement |
| Infinite Talk (480p) | 2,925+ | Photorealistic lipsync |
| Infinite Talk (720p) | 5,850+ | HD photorealistic lipsync |
Sound Effects (soundeffect-text)
| Model | Credits | Billing | Characteristics |
|---|---|---|---|
| ElevenLabs SFX | 0.78/s | perSecond | Text description → sound effect |
| Default | 2/s | perSecond | Basic sound effects |
| Stable Audio SFX | 78/track | Fixed | High-quality effect, 1 track |
TTS billed per character; no fixed list-display cost
Recommended Combinations by Use Case
| Goal | Recommended Model | Credits | Why |
|---|---|---|---|
| Quick idea sketching | Z-Image Free | FREE | Validate direction without spending credits |
| Social media illustration | Seedream 4.5 | 63 / 4 img | Illustration & anime sensibility, cost-effective |
| Portrait / influencer photos | Nano Banana 2 | 188 / 4 img | Natural skin tone and expression |
| Product photography | Flux 2 Pro | 50 / 4 img | Sharpness and color fidelity |
| Ad visuals | Nano Banana Pro | 234 / 4 img | Commercial quality, detailed finish |
| Text-in-image | GPT Image 2 (high) | 330 / 4 img | Industry-best text rendering |
| Korean-language prompts | GPT Image 2 / Nano Banana | 83+ / 62+ per 4 img | Only models with full Korean understanding |
| Video testing | LTX 2.3 Free | FREE | Explore video direction at no cost |
| Short social media video | Kling v2.1 Standard | 110 / 5 sec | Stable, cost-effective |
| Human motion video | Kling v3 Pro | 219 / 5 sec | Natural body movement |
| Korean-language video | Seedance 2.0 | 295 / 5 sec | Only video model with full Korean understanding |
| High-quality ad video | Seedance 2.0 | 295 / 5 sec | 720p, detailed motion |
| Native-audio video | Veo3.1 Fast | 624 / 8 sec | Audio generated natively alongside video |
| Background music | Lyria 3 | 15.6 / track | Natural sound, affordable |
Frequently Asked Questions
How big is the quality gap between free and paid models on xbrush?
It depends on the task. Z-Image Free and Hunyuan 3.0 Free produce results that are perfectly adequate for general social media content. For precise work — high-resolution portraits or advertising images with embedded text — paid models deliver a noticeable difference. A practical approach is to start with a free model to confirm the direction, then switch to a paid one if the output falls short.
How should I choose GPT Image 2's quality setting?
Use low (10 credits) for a rough preview, medium (83 credits) for standard content production, and high (330 credits) when text rendering accuracy or publication-grade quality is required. Because high costs four times more than medium, medium is usually sufficient for images that contain no text.
How much extra does it cost to add audio to a video?
It varies by model. The Veo3 series doubles in price when audio is included — for an 8-second clip, that means going from 624 to 1,248 credits. Kling v3 Pro rises about 50%, from 219 to 328 credits for 5 seconds. Unless you specifically need native audio generation, generating video without audio and adding music separately is the more cost-efficient approach.
Which models on xbrush truly understand Korean-language prompts?
Only a handful of models fully understand Korean: GPT Image 2, the Nano Banana series (Nano Banana, Nano Banana 2, Nano Banana Pro), and the Seedance series (Seedance 2.0, Seedance 2.0 Fast). All other models — including FLUX, Seedream, Qwen, Hunyuan, Wan, and SDXL — do not fully understand Korean. They may translate it internally or lose linguistic nuance. For Korean-language work, start with GPT Image 2 or the Nano Banana series.
What is the best value choice when credits are tight?
For images, start with Z-Image Free or Hunyuan 3.0 Free (no cost). For video, LTX 2.3 Free is the no-cost option. If you can invest a small number of credits, SDXL 1.0 (10 credits / 4 images) and Kling v2.1 Standard (110 credits / 5 seconds) offer reliable results at a low price.