Write Better AI Image Prompts: Formula, Examples, Fixes
A useful AI image prompt describes a scene someone could frame, rather than stacking praise words. Start with the five-part recipe below, copy a complete example, and change one variable at a time. Use CrePal's image creator to make a still, then review the image before animating it.
Why Your AI Image Prompts Keep Coming Out Generic
“Beautiful professional image” leaves the subject, setting, and light unresolved. A model may fill those gaps with familiar stock-image patterns. Compare: Weak: “A beautiful product photo.” Stronger: “A matte green ceramic mug on a pale oak table, side light from a north-facing window, 50 mm product photograph, quiet morning mood, plain background.” The stronger version makes the subject and composition testable. If the mug is wrong, revise its material or placement before adding more adjectives.
Second refinement pair: Weak: “A fantasy forest.” Stronger: “A narrow footbridge over a clear stream in an old beech forest, early morning mist, watercolor illustration, eye-level view from the bridge, quiet and hopeful mood.” If the first image loses the bridge, move “narrow footbridge” to the start and remove other competing objects. Compare the prompts and actual images separately; stronger wording is a clearer instruction, not proof of a better generated result.
Subject, Style, Light, Camera, Mood: The Five-Part Prompt Formula
Use subject + style + light + camera/composition + mood as a drafting checklist, not a guarantee that every model obeys every term. Subject answers what and where; style names the visual medium; light defines direction and quality; camera fixes framing; mood selects a coherent atmosphere.
Copyable template: “Create [subject and action] in [setting]. Render as [medium or visual style]. Light it with [source, direction, quality]. Frame it as [shot size or viewpoint]. Keep the mood [two compatible adjectives].” For an illustrated image, replace lens language with composition terms such as overhead view or centered silhouette.
Prompts to Copy for Portraits, Product Shots and Landscapes
Portrait: “An older violin maker seated at a workbench, scattered wood shavings behind her, editorial portrait, soft side light from one window, waist-up framing, patient and focused mood.”
Product: “A reusable steel water bottle standing on a damp slate slab, restrained studio product photograph, broad softbox from the left, three-quarter view, clean and outdoorsy mood.”
Landscape: “A narrow coastal path above a foggy cove at dawn, atmospheric landscape photograph, cool diffuse light, wide view with the path leading into the frame, calm and solitary mood.”
For each, first generate a draft. If the scene feels generic, change the subject detail or camera position, not all five parts at once. These are example inputs, not measured output comparisons.
Revise one visible problem at a time
| What went wrong | Wording to try | What to check |
|---|---|---|
| The mug is lost among props | “One matte green ceramic mug centered on an otherwise empty pale oak table.” | Is the mug the clear subject? |
| The composition is too wide | “Close product view of the mug, handle fully visible, with a small margin around it.” | Is the subject large enough without clipping the handle? |
| Reflections hide the surface | “Broad, soft side light; matte ceramic with subdued highlights.” | Can you read the material without distracting glare? |
Change one instruction and compare the next draft. This is a troubleshooting method, not a measured claim about output quality.
Got a Keeper? Turn the Image Into a Video Scene With CrePal
Choose the image with the clearest subject and fewest distracting artifacts. In CrePal's image-to-video tool, upload a mug image generated from the first example as the starting frame. Motion prompt: “Slow push toward the mug; steam curls upward; the mug and table remain still.” Preview edges, hands, text and continuity before using the clip in a larger project. Save the image prompt and motion prompt separately so you can revise either step without losing the other.
A Three-Draft Test for One Image Idea
Use the same subject for three drafts so each revision answers one question. Draft A: “A matte green ceramic mug on a pale oak table.” Draft B: add “soft side light from a north-facing window.” Draft C: keep that light and add “close three-quarter view, handle fully visible.” Save the prompt beside each image. Compare subject identity, light direction and framing separately; a nicer-looking image is not enough if the handle is cut off or the mug changes material.
If A misses the object, shorten the subject line before adding style. If B fixes the shadows but loses the green glaze, repeat the material cue near the subject. If C clips the handle, ask for a wider crop with margin around the object. Keep the best earlier draft rather than treating the newest one as automatically better. These are proposed test inputs and review criteria; the page does not claim a measured result from CrePal.
To turn a draft into a recurring world, carry the same subject into the fantasy prompt guide. For a painted version, use the watercolor palette and edge plan. If the result is meant to represent a brand, use the logo concept checks before treating it as a finished mark.
CrePal Product Sources and How to Use Them
CrePal's image creator says you can describe subject, setting, style and lighting, let the service select a model, or switch models manually. That makes a useful controlled comparison here: keep the subject and composition fixed while changing only the style or model. The page also documents taking an approved still into image-to-video; the one image run below documents a specific portrait prompt; other examples remain untested.
Related official reading: CrePal's product image guide. CrePal's product-image guide separates a clean, accurate hero image from more expressive lifestyle variants. For the bottle prompt above, check shape, material, label and included accessories against a real product reference before the image is used in a listing; a beautiful render is not evidence of product accuracy. The linked editorial guide supports the review checklist, not a measured result from the examples on this page.
Evidence and Review Notes
Evidence note added 5 November 2026. The prompts in this guide are editorial examples. The linked CrePal pages document product workflows or illustrative use cases; they do not verify an output from these exact prompts. Two prompt-output pairs are documented below; the other examples remain untested.
To document a result for this topic: Save the same mug prompt across three drafts and compare subject, material, light direction and handle visibility. Keep the exact prompt, model/version, settings, output URL, revisions and human review decision together.
First Dated CrePal Image Run
5 November 2026 · Seedream 5 Flash · 9:16 · 2K · Optimizer on · 1 image. The interface showed 5 credits for this run. The optimizer rewrote the input before generation; both versions are recorded below.
Original input: An older violin maker seated at a workbench, scattered wood shavings behind her, editorial portrait, soft side light from one window, waist-up framing, patient and focused mood.
Show the optimizer rewrite
An older violin maker seated at a workbench in three-quarter view, looking down with focused attention at a violin in her hands, waist-up framing, 50mm lens with the subject sharp and the cluttered workbench and wood shavings softly blurred in the background, soft side light from a single workshop window illuminating her face and hands, patient expression.
This run’s image appears in the cover above. Open the full-size R2 image.
Original output and test source
Original CrePal output. The cover image above is an R2 copy of that output.
Visible review: The output shows one older maker holding a violin at a tool-covered bench, with wood shavings and side light. These are visible matches to the scene prompt. They do not establish that another portrait prompt would keep the same person or remove the prop.
Evidence limit: This is the first of two account runs on this page. No blind reader assessment or named CrePal reviewer was involved; other prompts remain editorial examples.
Second run: inspect the Optimizer's prop change
5 November 2026 · Seedream 5 Flash · 16:9 · 2K · Optimizer on · 1 image · 5 credits shown. This second run tests one visible detail from the first example above.
Input: An older violin maker seated at a workbench, scattered wood shavings behind her, editorial portrait, soft side light from one window, waist-up framing, patient and focused mood.
Optimizer text and original CrePal output
An older violin maker seated at a workbench in three-quarter view, looking down with focused attention at the wood in her hands, scattered wood shavings behind her. Editorial portrait, waist-up framing, 50mm lens with the subject sharp and the cluttered workshop softly blurred in the background. Soft side light from one window illuminates her face and hands, creating a natural catchlight in her eyes.
Original CrePal output; the image below is the verified R2 copy.
Caption: Second CrePal image output · 16:9 · 5 November 2026.
What changed: The new wide image shows an older craftsperson at a wood-filled bench with side light, handling a piece of wood rather than the violin visible in the first image. The original input called her a 'violin maker' but did not require a visible violin. The Optimizer rewrite specified wood in her hands. If a violin must appear, name it as a required object and inspect the rewrite before generation.
Limit: The aspect ratio and/or Optimizer wording differ from the first run. These two images do not isolate one cause, establish repeatability or prove the other example prompts on this page.
Test your prompt in CrePal
Start with one example from this guide, then compare the result with your intended subject, style and details.
Site and support
CrePal's Terms identify SIGNOO PTE. LTD. as the service provider. Two dated CrePal runs are documented below; other prompt examples remain instructions to try.
Frequently Asked Questions
Long prompts with competing details get diluted. Lead with the subject and keep your style words consistent.
Some models support them to exclude unwanted elements, while others ignore them, so check your model's documentation.
Save the structure and swap only the subject and setting. Keep the seed or a reference image if your tool offers one.