As a professional creator working with high-resolution AI art (6,000 x 4,000 px), I’ve noticed a significant gap between ChatGPT’s image generation and other professional tools like Midjourney when it comes to texture rendering.
Currently, ChatGPT images often suffer from an ‘over-processed’ or ‘plastic’ look. Even at high resolutions, the micro-details in surfaces (skin pores, fabric fibers, weathered stone, organic noise) are missing or overly smoothed. This ‘uncanny valley’ of smoothness makes it difficult to use the outputs for large-format printing or professional-grade digital art.
I would like to suggest the following improvements:
Texture Frequency: Increase the level of high-frequency detail to avoid the ‘denoised’ effect.
‘Style Raw’ Mode: An option to bypass the heavy ‘AI-polish’ and get more organic, raw, and imperfect results.
Generative Upscaling: When requesting higher resolutions, the model should ‘hallucinate’ finer textures rather than just stretching existing pixels.
Consistency: Better preservation of tactile qualities across different prompts.
We need ChatGPT to move beyond ‘smooth renders’ into ‘photorealistic or tactile art’. High-resolution users need textures that feel real to the touch.
I really appreciate you taking the time to write this up. A lot of what you’re pointing at comes down to how current image models balance denoising, coherence, and detail. That tends to smooth out high-frequency texture (like pores, fibers, surface noise), especially as resolution increases. So even when the image is technically large, it doesn’t always feel detailed up close.
The ideas you listed especially a more “raw” rendering mode and better texture synthesis during upscaling are very much in line with the kind of feedback we’re seeing from advanced users.
I can’t promise timelines here, but I’ll make sure this gets captured properly as product feedback, this is exactly the level of specificity that helps.
Hi Smith. Another important issue for art: Persistent Character Seed & Spatial Consistency for Sequential Storytelling
I am a professional creator using ChatGPT for long-form visual projects, including graphic novels and storyboarding. While the creative capabilities of the model are impressive, there is a fundamental limitation preventing professional narrative use: the lack of character and environment consistency.
Currently, when generating a sequence of images featuring the same protagonist, the model fails to maintain physical traits (facial structure, specific tattoos, clothing details) or environmental layouts across different prompts. Each generation is treated as an isolated event, which is a deal-breaker for comic book creators or filmmakers who require visual continuity.
I would like to suggest the implementation of the following features:
Character Reference (CRef): The ability to assign a unique identifier or “seed” to a character’s design so the model can recall and replicate it in different poses, angles, and lighting conditions.
Persistent Environment Mapping: A way to “lock” a background or setting so that subsequent images maintain the same spatial logic and architectural details.
Global Style Seeds: An improved system to ensure that the artistic technique, texture, and color palette remain identical throughout a series of generations without “drifting” into different aesthetics.
To compete with tools that already offer “Character Reference” features, ChatGPT needs to allow us to build a visual “memory” for our projects. Without this, the tool remains a generator of “single beautiful images” rather than a professional instrument for sequential art.
Thank you for considering the needs of the narrative art community.