Together AI 上线 Black Forest Labs 的 FLUX.2 多参考图生图模型

• FLUX.2 on Together AI: Black Forest Labs' latest image model with multi-reference input for character/product consistency now available for 1M+ Together AI developers
• Key specs: Up to 4MP resolution, <10s generation, 32K characters, hex code color matching, works from 400x400px inputs
• Three models: FLUX.2 [dev], FLUX.2 [pro], FLUX.2 [flex]
• ‍Same SDK as LLMs: Drop-in compatible with existing Together AI code, serverless, or dedicated deployment

Image generation has cleared the demo bar, but production teams still can't trust it end to end. The misses aren't aesthetic, they're operational: brand colors land close but not exact, text degrades into noise, and characters or products subtly shift between shots, breaking continuity across a campaign or catalog. The result is a hidden tax of manual cleanup and rework that erases the promised speed. Even when a new model fixes one of these gaps, it typically comes as another standalone service with its own SDK, auth, billing, and limits — compounding fragmentation in stacks already running LLMs and voice at scale.

Today Together AI, the AI Native Cloud, is bringing FLUX.2 from Black Forest Labs to the Together Model Library, making production-grade image generation available for over 1 million AI developers. FLUX.2 targets the controls real applications require: multi-reference input for consistent characters and products across scenes, hex code color matching for strict brand compliance, and text rendering that holds up for typography, UI, and infographics. FLUX.2 [dev] is an open-weight model,  FLUX.2 [pro] ships as an optimized API model, and FLUX.2 [flex] offers tunable parameters, all served through Together AI's fast, reliable infrastructure and the same APIs used across the rest of the generative stack

Use cases

Character consistency across scenes

Game studios and content creators need the same character across different shots, poses, and lighting conditions. Multi-reference input locks in character identity while everything else changes.

Character Consistency — Base

Prompt: "Full body portrait of original fantasy character, young mage with silver hair, purple robes with gold embroidery, holding wooden staff, neutral gray background, concept art style, high detail"

Character Consistency — With Reference

Prompt with reference: "Same mage character from reference images, now casting spell with glowing hands, dramatic lighting from magic effects, ancient library setting, maintaining exact facial features and robe design, dynamic action pose"

Product design with color compliance

Custom products need exact color specifications maintained across different contexts and lighting. Hex code matching ensures brand colors hold.

Custom Product Consistency — Base

Prompt: "Minimal wireless speaker, geometric design, matte burgundy finish #8B1538, copper metallic accents, product photography on white background, soft studio lighting"

Product Consistency — With Reference

Prompt with reference: "Same speaker from reference on wooden desk in cozy reading nook, maintaining exact burgundy color #8B1538 and copper details, warm afternoon sunlight, lifestyle photography, books and plants nearby"

Brand identity across applications

Design teams building brand systems need visual consistency from logo to interface. Multi-reference input maintains design language while adapting to different use cases.

Logo / Brand Design — Base

Prompt: "Modern tech company logo, stylized lightning bolt icon, electric blue #00D9FF and deep purple #6B0FB3 color scheme, clean geometric design on white background, professional branding"

Logo / Brand Design — With Reference

Prompt with reference: "Same logo from reference applied to mobile app splash screen, maintaining exact colors #00D9FF and #6B0FB3, dark gradient background, glowing effect on icon, UI mockup"

Complex scene changes

Concept artists need characters that hold across dramatically different compositions, angles, and action states. Multi-reference input maintains identity through radical context shifts.

Character in Different Art Styles — Base

Prompt: "Cyberpunk character portrait, woman with neon pink mohawk, facial cybernetic implants, leather jacket with glowing circuit patterns, face close-up, photorealistic style, dramatic neon lighting"

Different Art Styles — With Reference

Prompt with reference: "Same character from reference, full body action shot, jumping between buildings in rainy cyberpunk city, maintaining exact facial features and cybernetics, motion blur, cinematic wide angle"

Technical specs

Platform integration

FLUX.2 runs on the same infrastructure handling your LLM and voice workloads. Same auth, same billing, same monitoring. Serverless APIs auto-scale through traffic spikes, dedicated endpoints guarantee isolated compute.

Infrastructure

• ✔ 99.9% uptime SLA
  

• ✔ North American data centers, SOC 2 Type II
  

• ✔ Auto-scaling serverless deployment
  

• ✔ <10s generation latency
  

• ✔ Real-time usage analytics
  

Developer Tooling

• ✔ OpenAI-compatible API patterns
  

• ✔ JSON structured prompting support
  

• ✔ Fine-tuning for FLUX.2 Dev
  

• ✔ Batch processing for high-volume workflows
  

• ✔ Same SDK as LLM endpoints
  

Code

Standard Together AI Python SDK, same patterns as text generation:

    
    from together import Together

    client = Together()

    response = client.images.generate(
        model="black-forest-labs/FLUX.2-pro",
        prompt="A mountain landscape at sunset with golden light reflecting on a calm lake",
        width=1024,
        height=768,
    )

    print(response.data[0].url)

    

Choose the model that fits your workflow: Dev for experimentation, Pro for production speed, Flex for text-heavy content.

Start building

Try FLUX.2 in Playground → Test multi-reference workflows, hex code matching, text rendering

Read the docs → API reference, structured prompting guide, optimization tips

Contact sales → Dedicated deployment, dedicated infrastructure, volume pricing