GPT Image 2.6: What's New, Flare vs Sunburst, and How to Prompt It

What is GPT Image 2.6 and when was it released?#

OpenAI officially released GPT Image 2.6 on November 4, 2026, rolling out the system globally across ChatGPT, Codex, and developer API endpoints.

The launch arrives as visual generation reaches unprecedented global scale, with OpenAI reporting that users now create more than 4 billion images per week across its platform ([OpenAI announcement](https://openai.com/index/introducing-images-2-6/)). That volume represents an average of over 500 million daily visual assets generated for marketing, concept design, and creative iteration. According to coverage by [The Next Web](https://thenextweb.com/news/chatgpt-animated-gifs-images-2-6-openai), the addition of animated GIF creation marks OpenAI's most notable motion capability rollout since the company retired Sora in March.

Continuing the model-family approach introduced with GPT Image 2.5, the 2.6 release provides two distinct developer engines: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. Flare is engineered for ultra-fast generation and structural layout quality, while Sunburst prioritizes micro-detail and cross-turn character retention. As noted in analysis by [Scalevise](https://scalevise.com/resources/chatgpt-images-2-6-gifs-reference-images-api-models/), this dual-model strategy allows enterprise teams to choose between turnaround latency and visual precision rather than accepting an arbitrary compromise.

What is new in GPT Image 2.6 compared with GPT Image 2.5?#

GPT Image 2.6 introduces several major upgrades over GPT Image 2.5: native animated GIF creation, updated Flare and Sunburst architectures, 6-second generation latency at low quality, an explicit complexity parameter, vector SVG export, and API-level image markup.

In GPT Image 2.5, creators could produce static raster frames but had to route images to external video models to introduce motion. With GPT Image 2.6, users in ChatGPT can prompt directly for short looping GIFs, inspect individual frames, and adjust playback speed on the fly. Furthermore, developers using the API can now pass coordinate-based image markup to pinpoint exact regions for edits, mirroring the visual brush tools previously restricted to the consumer ChatGPT web interface ([OpenAI System Card](https://deploymentsafety.openai.com/chatgpt-images-2-6/introduction)).

The table below contrasts the technical capabilities of GPT Image 2.5 against GPT Image 2.6 across every core parameter.

Sources as linked; checked November 5, 2026.

The introduction of the complexity parameter represents a significant workflow enhancement for prompt engineers. When set to low, the model skips intensive spatial reasoning, executing rapid color adjustments, background cutouts, and object swaps in seconds. For multi-character compositions or intricate packaging graphics, setting complexity to medium or high ensures the model plans out element hierarchies before generating pixels. For foundational background on prompting OpenAI image systems, consult our GPT image guide.

Flare or Sunburst: which model should you choose?#

You should choose GPT-Image-2.6 Flare for 6-second quick drafts, poster layouts, and social assets, while selecting GPT-Image-2.6 Sunburst when your creative task requires maximum visual fidelity and character consistency across multi-turn revisions.

OpenAI designed Flare to eliminate turnaround bottlenecks in high-volume creative operations. When quality is set to low in the API, Flare generates complete images in approximately 6 seconds, making it the preferred choice for real-time brainstorming, storyboard ideation, and rapid client mockups. It also delivers improved spatial layout balance for marketing posters, retail banners, and product graphics where clean typography placement outweighs micro-surface textures.

Sunburst functions as the precision engine in the GPT Image 2.6 suite. Configured at medium or high quality, Sunburst excels at retaining facial features, costume details, and brand identity across sequential editing turns. Cloud platforms like [Netlify](https://www.netlify.com/changelog/gpt-image-2-6-sunburst-flare-ai-gateway/) now provide zero-configuration access to both gpt-image-2.6-sunburst and gpt-image-2.6-flare, allowing teams to toggle between speed and quality inside serverless functions without maintaining separate infrastructure.

Sources as linked; checked November 5, 2026.

How do you make GIFs with GPT Image 2.6?#

You can generate animated GIFs in ChatGPT Images 2.6 by prompting directly for a looping action in chat or by requesting a 6-frame sprite sheet arranged in a precise 3-column by 2-row sequential grid.

Within the ChatGPT interface, making a GIF begins with a descriptive motion prompt, such as an expressive character reaction or a steam cycle rising from a warm beverage. Once the engine completes the asset, you can click on the graphic to play the loop, toggle playback speeds, or inspect individual frames to make targeted brush revisions ([The Next Web](https://thenextweb.com/news/chatgpt-animated-gifs-images-2-6-openai)). You can also upload an existing static photograph and prompt ChatGPT to animate specific ambient elements, such as drifting clouds, glowing neon signs, or flowing water.

For professional creators requiring precise frame control, OpenAI demonstrated a structured sprite-sheet prompting technique. This method instructs the model to generate a fixed grid of identical panels, maintaining a locked camera position, uniform lighting, and consistent props while advancing character motion incrementally across frames. To ensure an infinite loop, the motion in the final frame must transition seamlessly back into the pose of the first frame.

Use the two original, copy-ready prompts below to test this sprite-sheet looping technique in your own workflow:

Create a single animation sprite sheet for a seamless looping GIF. An isometric miniature scene of an android barista pouring glowing cyan matcha tea into a ceramic cup on a weathered wooden counter. The sequence: barista lifts copper kettle, tilts spout, glowing liquid streams into the cup creating a swirling vortex, steam forms miniature holographic rings, barista lowers kettle, stream thins and breaks smoothly as steam rings fade. Exactly 6 sequential frames arranged in a precise 3-column by 2-row grid, read left-to-right then top-to-bottom. Every panel has identical dimensions. Fixed 45-degree isometric camera angle, same android character, kettle, cup, counter, and warm neon-lit coffee stall background. Clear incremental liquid and steam motion; sixth frame transitions smoothly toward the first. Stylized 3D miniature claymation aesthetic, cozy cinematic rim lighting, gentle futuristic atmosphere. No captions, no words, no numbers, no logos, no panel labels, no borders, no gutters between panels.

Create a single animation sprite sheet for a seamless looping GIF. A miniature yellow robotic bumblebee drone pollinating a glowing glass orchid inside a high-tech greenhouse terrarium. The sequence: drone hovers with wings fluttering, descends gently toward the open flower blossom, taps petal releasing a burst of glittering golden pollen motes, hovers in place while collecting glowing nectar, begins ascending upward, reaches peak hovering altitude ready to descend. Exactly 6 sequential frames arranged in a precise 3-column by 2-row grid, read left-to-right then top-to-bottom. Every panel has identical dimensions. Fixed macro side-profile camera angle, same drone chassis, flower stem, glass dome, and soft blurred greenhouse background. Clear incremental wing blur, altitude shifts, and floating pollen particles; sixth frame transitions smoothly toward the first. Detailed macro digital art style, soft bioluminescent pastel lighting, playful mechanical rhythm. No text, no captions, no panel borders, no speech bubbles, no gutters.

How do you write better GPT Image 2.6 prompts?#

Writing effective prompts for GPT Image 2.6 requires structuring your instructions into 4 core pillars: subject definition, spatial composition, visual style, and explicit modification constraints that specify what must remain unchanged.

Generative models frequently fail when given vague aesthetic descriptors like "stunning" or "photorealistic." GPT Image 2.6 responds with significantly greater accuracy when you specify concrete camera focal lengths, lighting origins, and physical material properties. When executing multi-turn revisions, OpenAI's prompting guide emphasizes stating explicitly which elements must stay identical and which specific attributes to modify, preventing the model from altering established character traits or lighting environments.

Review the good versus bad prompting comparison below to identify common structural errors and understand why optimized instructions yield superior visual consistency:

To help you get started immediately, here are 4 production-grade copyable prompts crafted for the GPT Image 2.6 architecture:

Comprehensive character reference sheet for a fantasy archer. Full-body orthographic turnaround showing front view, three-quarter view, and rear profile side by side on a neutral light-gray background. Subject: 28-year-old female ranger with sharp angular features, athletic build, high cheekbones, green eyes, and silver hair tied in a practical braided bun. Wardrobe: fitted forest-green leather jerkin with embossed leaf patterns, dark charcoal wool trousers, reinforced bracers on forearms, and knee-high soft leather boots. Equipment: recurve yew bow slung across torso and quiver of white-fletched arrows on right hip. Style: detailed digital concept art, clean line work, realistic fabric folds and leather weathering, neutral studio lighting with soft ambient occlusion shadows. No background props, no text labels, no watermarks, no borders.

Commercial studio hero shot of a luxury matte-black ceramic pour-over coffee dripper resting atop a ribbed borosilicate glass carafe. The assembly sits centered on a raw gray slate slab surrounded by a light dusting of coarse ground coffee beans. Composition: eye-level studio macro shot with an 85mm lens perspective, tight focus on the dripping coffee droplet suspended at the dripper base, soft atmospheric depth of field blurring the background. Lighting: soft diffused overhead key light with an amber back-rim light accentuating the transparency of the glass carafe and rising steam. Style: premium editorial product photography, ultra-clean surface reflections, true-to-life ceramic matte texture. Constraints: no brand logos, no text markings, no extraneous kitchen utensils.

Minimalist architectural exhibition poster celebrating brutalist concrete design. Composition: bold asymmetrical layout with the upper two-thirds occupied by a stark black-and-white photograph of a spiraling cast-concrete staircase viewed from below; the lower third provides clean negative cream space framed by sharp geometric hairline rules. Color palette: deep charcoal, raw concrete gray, and warm off-white with a single vivid vermilion red accent square in the upper left corner. Style: Swiss international typographic style, sharp structural edges, authentic analog film grain, high architectural contrast. Constraints: no placeholder Latin text, no distorted letters, no clutter, pristine editorial layout balance.

Cinematic storyboard production frame for a suspense thriller sequence. Camera composition: intense Dutch-angle medium close-up from inside a dimly lit subway car looking toward the sliding double doors. Subject: exhausted male detective in a damp beige trench coat gripping the chrome handrail, eyes darting toward the dark tunnel window reflection where a shadowy figure stands in the adjacent car. Lighting: flickering fluorescent overhead tube lights casting harsh green-tinted shadows, punctuated by periodic warm amber flashes from passing platform signals. Style: anamorphic widescreen 2.39:1 aspect ratio film still, realistic film grain, subtle motion blur on passing tunnel pillars, gritty neo-noir atmosphere. Constraints: no stylized cartoon filters, authentic 35mm film color grading.

If your project requires dedicated character consistency across full comic books or animation bibles, review our AI character generator walkthrough and our reusable storyboard template for production pipelines.

Can you use GPT Image 2.6 on DramaCanvas?#

GPT Image 2.6 isn't on DramaCanvas yet (checked November 5, 2026); we'll add it when it becomes available to us. Until then, GPT Image 2.5 and GPT Image 2 are on the canvas. Both currently supported engines offer robust multi-reference generation, precise aspect ratio controls, and transparent background outputs. You can orchestrate end-to-end creative campaigns today on our infinite canvas, combining static generation with downstream animation tools.

How do you make your first GPT Image picture on DramaCanvas?

Add an image node, choose GPT Image 2.5, set the ratio and resolution, write a prompt and generate.

Step 1: Add an image node and choose GPT Image 2.5

Click Add node in the toolbar at the bottom of the canvas and choose Image. In the model picker under the node, choose GPT Image 2.5.

Step 2: Pick the ratio and resolution in Image settings

Open Image settings from the settings summary below the prompt (for example 1:1 · 1K) and choose the ratio under Size or aspect ratio: 1:1 for avatars and product main images, 4:5 or 3:4 for Instagram posts, 9:16 for vertical covers, 16:9 for banners. Start with Resolution at 1K. If you need a cut-out asset, turn on Transparent background.

Step 3 (optional): Connect reference images

To keep a particular person, product or style, add the reference image to the canvas and drag a line from the connection point on the right of its node to your new image node. In the prompt, type @ and pick Image 1 from the menu, then say how the model should use it. You can connect up to 8 images.

Step 4: Write the prompt, check the credits, generate

Paste one of the prompts below into the input box and adapt it. The generate button shows how many credits this image will cost; check it, then click. When it's done, save it with the download button on the node:

• Character concept: Fantasy archer in forest-green leather jerkin, neutral background, soft studio lighting.
• Product hero: Matte-black ceramic pour-over dripper on slate slab, macro lens, warm rim lighting.
• Graphic poster: Minimalist brutalist architectural composition, sharp contrast, clean editorial negative space.

The table below outlines current model capabilities and credit consumption across image generators on DramaCanvas:

DramaCanvas Standard plan provides 1,100 credits for $29.90 per month (approximately $0.027 per credit). Verified November 5, 2026.

Once your image finishes generating on DramaCanvas, you can immediately animate it into cinematic footage without waiting for native GIF support. Drag a connection line from your image node to a video node, open the video parameters, and select your graphic under Parameters → Start frame. You can route images through cutting-edge engines such as Kling 3.0, Kling 4, or Seedance. For step-by-step guidance, follow our image to video guide, explore our review of the best AI video generators, or read our Nano Banana Pro guide.

How do you use GPT Image 2.6's new ideas on the DramaCanvas infinite canvas today?#

You can put GPT Image 2.6's core creative workflows into practice right now on DramaCanvas using native canvas nodes and GPT Image 2.5:

1. Character turnaround sheets: Generate a character reference sheet with GPT Image 2.5, then connect that node directly to several video nodes to lock identity and wardrobe across consistent shots ([[link:ai-character-generator

   AI character generator]]).

2. Motion and animation loops: Instead of static sprite sheets, generate a key pose image and animate it with a video model via Parameters → Start frame, or connect two distinct poses as Start and End frames on Kling 3.0 ([[link:image-to-video-ai

   image to video guide]]).

3. Targeted iterative edits

   Practice the "edit only what must change" principle by connecting your previous image node as a reference (@Image 1) and specifying in the prompt exactly what stays unchanged.

4. Storyboard frame sequences: Use the Storyboard script node to convert a narrative outline into structured shot rows, then generate all storyboard frame images directly on the canvas ([[link:storyboard-template

   storyboard template]]).

5. High-resolution posters and cutouts

   Create marketing posters and product staging layouts with GPT Image 2.5 at 2K or 4K resolution, enabling Transparent background for instant cutout compositing.

What are the safety safeguards and limitations of GPT Image 2.6?#

OpenAI's official system card published on November 4, 2026, details a multi-tiered safety pipeline evaluated against 3,077 adversarial test prompts.

The safety stack operates through two sequential evaluation layers: upstream LLM-based policy checks and a downstream multimodal safety reasoning monitor ([OpenAI System Card](https://deploymentsafety.openai.com/chatgpt-images-2-6/introduction)). Upstream checks analyze text prompts before generation begins, immediately refusing violative requests. Downstream, the safety reasoning monitor moderates both the input reference imagery and the final generated output, while synchronous Child Sexual Abuse Material (CSAM) filters guard against illicit content across both still and animated formats.

The table below summarizes OpenAI's automated evaluation metrics across 3,077 adversarial test cases, comparing GPT-Image-2.6 against the previous GPT-Image-2.5 baseline:

Data sourced from OpenAI Deployment Safety Hub, published November 4, 2026.

Despite these rigorous safeguards, creative teams must account for clear operational boundaries when deploying GPT Image 2.6. The model cannot generate long-form continuous video files, remaining restricted to short looping GIFs and sprite sheets. Furthermore, while vector SVG export works well for clean icons and geometric layouts, intricate photographic scenes still degrade when forced into vector formats. For automated video production workflows beyond simple GIFs, you can also explore our Claude video generation MCP integration.