Hello all,
I am bit struggling with the spelling errors and text rendering issues with the image gen model (gpt-image-1). My understaning is that the newer model is specifically good at this. Are there any specific instructions that can help to avoid such errors?
Here is my detailed prompt example and the output
Graphical Abstract Design Instructions: T-ALL Epigenetic Subtypes and 5-Azacytidine Response
Overall Layout
Create a horizontally-flowing scientific infographic with white background, using a cohesive color scheme where each T-ALL subtype (C1-C5) has its own distinct color maintained throughout all panels. Use professional scientific styling with clean lines (2-3px thickness) and high resolution (1200+ DPI). Organize content in 3 connected panels flowing left to right.
Panel 1: T-ALL DNA Methylation Classification
Create a DNA double helix backbone at the top with CpG sites represented as small circles on the strand
Below this, show 5 distinct T-ALL methylation profiles arranged horizontally (C1-C5):
For C1: Show sparse methyl groups (red circles) representing hypomethylation
For C2-C4: Show moderate patterns of methylation
For C5: Show dense methyl groups representing hypermethylation
Label each profile with only: "C1", "C2", "C3", "C4", and "C5"
Under each profile, include small simplified representations of the associated genetic features:
C1: Two interlinked protein structures labeled "DNMT3A/IDH2"
C2: A transcription factor binding to DNA labeled "TAL1"
C3: A protein-DNA interaction labeled "TLX3"
C4: A protein-DNA interaction labeled "TLX1/HOXA9"
C5: A protein with upward arrow labeled "HOXA9"
Panel 2: Epigenetic Heterogeneity and Clinical Outcomes
Create a split visualization for C1 (hypomethylated) and C5 (hypermethylated) subtypes
For C1: Show T-cells (simplified circular cells with T-cell receptors) with sparse methyl marks and an arrow pointing to text "Poor outcome"
For C5: Show T-cells with dense methyl marks and an arrow pointing to text "Poor outcome + treatment resistance"
Include a small Kaplan-Meier-style curve showing lower survival probability for C1 and C5 compared to other subtypes
Panel 3: 5-Azacytidine Treatment Response
Create a mouse xenograft visualization showing:
A simplified mouse silhouette with human T-ALL cell implantation site
Split the visual to show untreated vs. treated conditions
For untreated: Show hypermethylated cells (C5) proliferating extensively
For 5-azacytidine treated: Show cells with reduced methylation and reduced proliferation
Include a small Kaplan-Meier-style curve showing improved survival for treated C3, C4, and C5 samples
Add a molecular structure icon of 5-azacytidine with label "5-azacytidine"
Use dotted arrows to show the progression from treatment to demethylation to improved survival
Connecting Elements
Use consistent 2.5px arrows to show flow between panels
Connect DNA methylation subtypes to their clinical characteristics
Use color-coding consistently for each subtype (C1-C5) throughout all visualizations
Add a final arrow leading from Panel 3 to a small representation of a medication bottle labeled "Epigenetic therapy"
Technical Specifications
Maintain scientific accuracy in all molecular representations
Use semi-transparent coloring for cellular elements
Ensure all text is minimal, only using labels specified in quotation marks
Present data visualizations in a simplified but scientifically accurate manner
Use consistent line weights: 3px for primary flows, 2px for secondary elements
Apply subtle drop shadows only to separate overlapping elements if necessary
CRITICAL:
- No image title or conclusion texts
- All the content should be strictly organized within the canvas size of 1024x1024 PNG
- Use as minimal amount of text as possible and restrict it only to labels
- Render all text using clean, vector-based fonts suitable for publication or scientific presentations. Avoid blurry, pixelated, or handwritten-style text.
- Strictly no spelling errors! Ensure all text labels are spelled exactly as provided in the prompt. Do not guess, auto-correct, or generate new scientific terms.
- Make absolutely sure that texts are rendered to perfection. It should not be swiggly or curved. They should be readable.
- Contents must be crisp and well organised
- Use scientifically and anatomically accurate icons to represent the graphical elements
- Overall output should be perfect and be able to be readily submitted to a high-impact Cell Press journal for publication
Although it gets most of the stuff correctly, it does makes mistakes as in bottom left corner.
Any tips would be appreciated. Thank you 
