AI Image Prompt Guide: Text-to-Image, i2i Edits and LoRA Prompts
How to write prompts for the image models in the Imagera Sandbox, with tested examples, the controls that matter and honest limitations
A text-to-image prompt describes a picture. An i2i (image-to-image) prompt is an instruction about a photo you already have. A LoRA prompt leaves the look to the LoRA. This guide covers all three, for Imagera Typeset 3, the edit models (Imagera Typeset 3 (Edit), Fast Edit, Precise Edit) and the LoRA image generator.
Why the Prompt Type Matters
Most prompt guides treat every request the same. In practice there are three kinds: describing a new image, instructing an edit to an existing one, and prompting on top of a LoRA. Each needs different wording. The Sandbox handles the technical settings for you, so what is left to get right is the prompt, the frame shape and the output size.
Step-by-Step Guide
Start with a Clear, Specific Subject
Begin every prompt with your main subject. Be specific about who or what you want to see. "A woman" gives the AI too many options. "A professional businesswoman in her 30s with auburn hair and glasses" gives it a clear target. The more specific your subject, the fewer reruns you'll need.
Pro Tips
- • Use specific descriptors: age, features, expression, pose
- • Name the subject's action: "standing", "walking through", "looking at camera"
- • Include distinguishing details: "wearing a navy blazer", "holding a coffee cup"
- • For objects: specify material, size, condition — "a weathered leather journal" vs "a book"
Define the Setting and Environment
The environment provides context, mood, and atmosphere. "Modern Tokyo street at night with neon reflections on wet pavement" tells the AI exactly what you want. "City" leaves everything to chance. Include location, time of day, weather, and key environmental elements.
Pro Tips
- • Be specific about location: "a sunlit Italian cafe terrace" vs "outside"
- • Include atmospheric details: fog, rain, golden hour light, overcast sky
- • Describe the relationship between subject and environment
- • Add depth cues: foreground elements, background details, middle ground
Specify the Visual Style
Style is the single most impactful element after subject. Without a style directive, the model falls back on its own default look. Name the style: "editorial fashion photography", "cinematic film still", "hand-drawn anime", "dark fantasy digital painting". The more specific, the more consistent.
Pro Tips
- • Photography styles: "editorial", "street photography", "product shot", "portrait"
- • Art styles: "oil painting", "watercolor", "concept art", "digital illustration"
- • Describe the aesthetic: "symmetrical pastel color palette", "cyberpunk neon"
- • Combine styles for unique looks: "cinematic portrait with film grain and muted tones"
Add Lighting and Camera Details
Lighting separates amateur prompts from professional ones. "Golden hour backlighting with rim light on hair" produces a very different result from no lighting instruction at all. For photorealistic work, camera details help too: lens type affects perspective, aperture affects depth of field.
Pro Tips
- • Lighting keywords: "golden hour", "rim lighting", "studio softbox", "dramatic shadows", "Rembrandt lighting"
- • Lighting fades: "light fading from warm to cool across the frame", "soft falloff into shadow"
- • Camera lens: "85mm portrait lens" (compressed), "35mm" (natural), "wide-angle" (dramatic)
- • Depth of field and angle: "shallow depth of field", "low angle hero shot", "overhead flat lay"
Use Composition and Mood Keywords
Stating the composition stops the model defaulting to a centered subject. Mood keywords steer color, atmosphere and emotional tone. These small additions make the difference between a technically correct image and one that feels intentional.
Pro Tips
- • Composition: "rule of thirds", "negative space on the left", "leading lines", "symmetrical"
- • Mood: "melancholic", "energetic", "peaceful", "tense", "whimsical"
- • Color: "warm tones", "desaturated", "high contrast", "pastel palette"
- • Texture: "grainy film texture", "smooth gradient", "sharp details"
Say What You Want Instead of What to Avoid
The Sandbox image models take one prompt; there is no separate negative-prompt box. Instruction-following models do best with positive wording: "a clean white background" works better than "no clutter". Keep any exclusions short and put them at the end of the prompt.
Pro Tips
- • Rephrase exclusions as the thing you want: "bare wall" rather than "no posters"
- • Short exclusions at the end are fine: "no text, no watermark"
- • For anatomy, describe the pose clearly: "hands resting on the table, fingers relaxed"
- • If one detail keeps going wrong, fix that detail with an edit instead of rewriting the whole prompt
The Prompt Formula
Use this structure for text-to-image prompts: [Subject] + [Setting] + [Style] + [Lighting] + [Composition/Mood] + [Details]. Not every prompt needs all six parts. A product shot might skip mood, and a portrait might skip the environment. Covering at least Subject, Style and Lighting gives reliably good results.
- Subject: Who or what — be specific with descriptors
- Setting: Where and when — location, time, atmosphere
- Style: Artistic approach — photography type, art style, aesthetic
- Lighting: Light source, direction, quality — the biggest single improvement
- Composition/Mood: Arrangement, emotion, color palette
- Details: Lens, materials, texture, any exact text in quotes
Text-to-Image Prompts — Imagera Typeset 3
Imagera Typeset 3 is the Sandbox's default image model, built for readable lettering in posters, signs and layouts. It follows long, natural-language descriptions well, so write full sentences rather than keyword lists.
- Put the exact words you want in the image in double quotes: a sign reading "OPEN LATE"
- Say where the text goes and how it looks: "bold white sans-serif title across the top"
- Aspect pill: pick the frame shape before you run (vertical for phones, wide for screens)
- Resolution (Advanced): 1K or 2K. Output size is usually the biggest single lever on cost
- Speed: Normal or Ultra Fast. Ultra Fast changes the sizes on offer and the price shown
- Format (Advanced): PNG or JPEG
- Imagera Typeset 3: from 15 credits per image at the default size. Larger sizes cost more. The exact cost shows on the Generate button before you run.
Image-to-Image (i2i) Prompts — The Edit Models
An i2i prompt is an instruction, not a description. You upload a photo and say what to change: "Change the background to a sunset beach" or "Make the dress red". Each edit model takes the photo plus optional reference images, for example an outfit or a product to place into the scene.
- Imagera Typeset 3 (Edit): changes the frame while keeping lettering readable. Good for signs, packaging and text swaps
- Fast Edit: one short sentence, fast result. Good for quick outfit, hair and background changes
- Precise Edit: long, multi-part briefs. Offers 1K, 2K and 4K and a transparent-background option
- Write instructions: "Change X to Y", "Add Z to the table", "Remove the car on the left"
- Say what to KEEP as well as what to change: "keep the pose and face, change the outfit"
- Directions are from the viewer's side: "move the lamp to the left" means your left
- For big changes, work in steps: structure first, then details
- Set the Aspect pill to your photo's shape (or auto where offered), or the frame will be recomposed
- Imagera Typeset 3 (Edit): from 15 credits per edit; Fast Edit: from 15 credits per edit; Precise Edit: from 15 credits per edit. The exact cost shows on the Generate button before you run.
7 Edit Presets With Ready-Made Instructions
These tools open Imagera Typeset 3 (Edit) with a starter instruction for one job. Replace the bracketed part with your own words:
- AI Background Studio: "Replace the background with [scene description]" — product shots, portraits
- AI Angle Changer: "Show this person from a 45-degree side profile" — rotates the subject
- AI Look Book: "Change the outfit color to emerald green" — recolors and swaps clothing
- Photo Style Transfer: "Turn this photo into a hand-drawn anime illustration" — applies an art style
- AI Product Photography: "Place this product on a marble countertop with soft studio lighting"
- Smart Detail Enhancer: "Remove flyaway hair and clean up skin details" — fine detail refinement
- AI Text Editor: "Change the sign text to OPEN" — rewrites text while keeping the font style
LoRA Prompting — How Custom Styles Change Your Prompt
A LoRA (Low-Rank Adaptation) is a small trained add-on that changes how the base model draws: a photorealistic LoRA makes everything more photo-like, a style LoRA applies one consistent aesthetic, a character LoRA keeps one face. Once a LoRA sets the look, your prompt should describe the content.
- LoRAs handle STYLE — your prompt should focus on CONTENT (subject, scene, composition)
- Don't fight the LoRA: with a photorealistic LoRA attached, don't prompt for "anime style"
- Strength runs from 0 to 2 and starts at 0.8. Lower it if the LoRA overpowers the prompt
- Up to 3 LoRAs can be chained — give each its own strength to balance them
- LoRA sources: the built-in style library, your own upload or link, or a LoRA you train (CivitAI links are not supported)
- Check trigger words — many LoRAs need a specific keyword in the prompt to activate
- LoRA image generator: from 20 credits per image, plus 5 credits for each LoRA you attach. The exact cost shows on the Generate button before you run.
What AI Image Models Are NOT Good At — Honest Limitations
No image model is perfect. Knowing the common failure points saves reruns and credits:
- Hands and fingers: still the most common failure — extra or merged fingers, impossible poses. Describe the hand pose plainly, and fix a bad hand with an edit rather than a full rerun.
- Long text: Imagera Typeset 3 renders short text well, but long sentences, tiny type and unusual fonts can still break. Keep text short and in quotes.
- Counting: "exactly 3 birds" may give 2, 4 or 5. Precise quantities are unreliable.
- Crowded scenes: past 2-3 people, errors rise — merged bodies, extra limbs, mismatched lighting.
- Teeth and ears: often too small, crowded or asymmetrical in close-ups. "Natural teeth, detailed ears" helps.
- Physics and shadows: shadows can point the wrong way and objects can float.
- Complex poses with objects: "a handstand while holding a cup" will likely fail.
- Repeated edits: each edit of an edit can lose a little quality. Make big changes in as few passes as possible.
Edit Models — Known Limitations
Edit models are powerful, but know these before you spend credits:
- Unrequested changes: an edit can touch parts you meant to keep. Name them: "keep the background exactly as is".
- Reframing: if the Aspect pill does not match your photo, the output is recomposed to the new shape.
- Identity drift: faces can shift slightly. Say "keep the face and expression unchanged" when it matters.
- Hard lettering: unusual characters may not come out right in one pass. Edit the text in its own step.
- Progressive degradation: repeated edits on the same image reduce quality. Plan to use few passes.
15 AI Image Prompt Examples That Work
Starting prompts for each kind of request. Copy one, then change the subject and details:
- Imagera Typeset 3 — Portrait: "Professional headshot of a confident woman in her 40s, silver hair, minimal makeup, navy blazer, neutral gray studio background, soft diffused lighting, 85mm lens, shallow depth of field, sharp focus"
- Imagera Typeset 3 — Landscape: "Misty mountain valley at sunrise, pine forest in the foreground, golden light breaking through clouds, volumetric fog, wide-angle lens, nature documentary photography"
- Imagera Typeset 3 — Product: "Premium wireless headphones on a dark slate surface, dramatic side lighting, soft gradient background, product photography, macro detail on the materials, clean commercial look"
- Imagera Typeset 3 — Fantasy: "Ancient library with floating books and glowing runes, warm candlelight, dust in the light beams, concept art style, rich warm color palette, detailed environment"
- Imagera Typeset 3 — Poster text: "Minimal travel poster of a lighthouse at dusk, flat vector style, the title \"NORTH COAST\" in bold white capitals across the top, muted teal and orange palette"
- Imagera Typeset 3 — Storefront sign: "Modern coffee shop storefront with a neon sign reading \"OPEN LATE\", brick exterior, warm glow through the windows, rainy evening, reflections on the wet sidewalk"
- Imagera Typeset 3 — Street photo: "Documentary photograph of a street food vendor at night, steam rising from the wok, warm tungsten light mixed with blue neon, candid moment, 35mm film grain"
- Imagera Typeset 3 — Fine art: "Ethereal underwater portrait of a dancer, flowing fabric forming abstract shapes, dappled light from above, deep blue-green palette, fine art photography, dreamlike"
- Imagera Typeset 3 — Macro: "Macro photograph of morning dew on a spider web, each droplet reflecting the sunrise, extreme close-up, soft bokeh of a wildflower meadow, natural light"
- LoRA image generator: "Woman in an elegant evening gown on a rooftop terrace at sunset, city skyline behind her, wind in her hair, golden hour backlight" (with a photorealistic LoRA from the style library)
- Fast Edit — Background (i2i): "Replace the background with a tropical beach at sunset with palm trees and a calm ocean" (input: portrait photo)
- Imagera Typeset 3 (Edit) — Style transfer (i2i): "Turn this photo into a hand-drawn anime illustration, keep the original composition and poses" (input: group photo)
- Fast Edit — Outfit change (i2i): "Change the outfit to a black tuxedo with a bow tie, keep the same pose and expression" (input: casual photo)
- Imagera Typeset 3 (Edit) — Text edit (i2i): "Change the text on the sign to \"IMAGERA\" in the same font style and color" (input: photo with a sign)
- Precise Edit — Product staging (i2i): "Place this product on a clean white marble surface with soft studio lighting and a subtle shadow, keep the label sharp and unchanged" (input: product photo)
Quick Controls Reference
What you set outside the prompt, and when to change it:
- Model: pick by job — new image, text in the image, quick edit, detailed edit, or a LoRA style
- Aspect: the frame shape. Set it before you run; the prompt does not change it
- Resolution: in Advanced. Imagera Typeset 3 offers 1K and 2K; Precise Edit goes up to 4K. Bigger costs more
- Speed: Normal or Ultra Fast. The price on the Generate button follows your choice
- Reference images (edit models): add them to show an outfit, product or style to use
- LoRA strength: starts at 0.8. Raise it for a stronger look, lower it for subtler
- The exact cost shows on the Generate button before you run.
Common Mistakes That Waste Credits
The most frequent prompting mistakes, each of which costs a rerun:
- Being too vague: "pretty landscape" gives unpredictable results — add location, time, style
- Overloading prompts: 20 ideas in one prompt confuse the model — focus on 3-5 key elements
- Conflicting instructions: "bright sunny night scene" — the model can't resolve contradictions
- Wrong prompt type: describing a whole new picture to an edit model, or giving instructions to a text-to-image model
- Asking the prompt for a size: "make it 4K" does nothing — use the Resolution control
- Mismatched Aspect on an edit: a portrait photo on a wide frame gets recomposed
- Expecting perfect long text: short text in quotes works; paragraphs do not
- Fighting the LoRA: prompting for a style that contradicts the attached LoRA
Expert Tips for Better Results
Frequently Asked Questions
How long should my AI image prompt be?
For text-to-image: usually 30-100 words. Cover subject, style, lighting and composition without padding. Imagera Typeset 3 follows full sentences, so there is no need to write keyword lists. For i2i edits: one or two short sentences that say what to change and what to keep. Very long prompts dilute attention and often do worse than concise, specific ones.
Which Imagera model should I use for my prompt?
For a new image, especially one with text in it, use Imagera Typeset 3. To change a photo you already have, use an edit model: Fast Edit for a quick one-sentence change, Imagera Typeset 3 (Edit) when lettering must stay readable, and Precise Edit for a detailed brief or 4K output. For a consistent custom style or character, use the LoRA image generator with a LoRA attached.
What are i2i prompts, and how are they different from text-to-image prompts?
An i2i (image-to-image) prompt edits a picture you upload; a text-to-image prompt creates one from nothing. Text-to-image prompts describe the result ("a red dress on a beach at sunset"). i2i prompts are instructions ("change the dress to red, keep the pose and background"). Imagera Typeset 3: from 15 credits per image at the default size; Imagera Typeset 3 (Edit): from 15 credits per edit. The exact cost shows on the Generate button before you run.
Can a reference image change the size of the generated image?
No. A reference image guides content and style: the subject, an outfit, a product, a look. The output size comes from the Aspect control (the frame shape) and the Resolution control (the pixel size), which you set separately. Writing "make it bigger" or "4K" in the prompt does not change the size either.
Why do my AI-generated images have weird hands?
Hands are the hardest thing for any image model: the anatomy is complex and changes with every angle. To reduce problems, describe the hand pose plainly ("hands resting on the table, fingers relaxed"), avoid poses where hands are the focal point, and keep scenes to a few people. If one hand still comes out wrong, fix it with an edit model instead of rerunning the whole image. Expect occasional failures; this is a limitation of all current models.
Can I put text in my AI-generated images?
Yes. Imagera Typeset 3 is built for readable lettering in posters, signs and layouts. Put the exact words in double quotes (e.g., "OPEN"), keep them short, and say where the text goes and what it looks like. To change text in an existing photo, use Imagera Typeset 3 (Edit), which keeps lettering readable while editing the rest. Long paragraphs are still unreliable on every model.
What are LoRAs and how do I use them with prompts?
LoRAs are small trained add-ons that change the base model's style, character or look. They handle STYLE while your prompt handles CONTENT, so focus the prompt on subject, scene and composition. Check the LoRA's trigger words, since many need a specific keyword. Strength starts at 0.8 and runs from 0 to 2. You can chain up to 3 LoRAs at once, from the built-in style library, your own upload or link, or a LoRA you train. LoRA image generator: from 20 credits per image, plus 5 credits for each LoRA you attach.
How can I make my AI images more photorealistic?
Use photography language: the camera ("DSLR", "mirrorless"), the lens ("85mm f/1.4"), the light ("natural window light", "golden hour") and the texture ("realistic skin texture, natural pores"). Describe an ordinary, candid moment rather than a perfect one. Attach a photorealistic LoRA from the Realistic & Photography section of the style library, and raise Resolution for more detail.
Why does the edit change parts of my image I didn't ask it to?
Edit models redraw the image, so parts you meant to keep can shift. To reduce this, name what must stay the same ("keep the background, lighting and expression exactly as is, only change the outfit"), set the Aspect control to match your photo, and make one targeted change per edit rather than several at once.
What settings should I use for the best results?
The Sandbox sets the technical parameters for each model, so there are no steps or CFG values to tune. What you choose is the model, the Aspect (frame shape), the Resolution (in Advanced) and the Speed. Test at the default size, then rerun your best prompt larger. For LoRAs, start at the default strength of 0.8 and adjust. The exact cost shows on the Generate button before you run.
Why is my AI image low quality or distorted, and how do I fix it?
Low detail: raise the Resolution, and add lighting and texture words to the prompt. Distorted anatomy: simplify the pose and the number of people, then fix remaining faults with an edit. Wrong style: name the style explicitly, or check that an attached LoRA isn't fighting the prompt. An edit that ignores your instruction: make it shorter and more specific, and say what to keep. Stretched or recomposed edits: set Aspect to your photo's shape.
How much does each AI image cost on Imagera?
Imagera Typeset 3: from 15 credits per image at the default size. Imagera Typeset 3 (Edit): from 15 credits per edit; Fast Edit: from 15 credits per edit; Precise Edit: from 15 credits per edit. LoRA image generator: from 20 credits per image, plus 5 credits for each LoRA you attach. Larger sizes and the Ultra Fast route can cost more. The exact cost shows on the Generate button before you run.
Related Guides
LoRA Training Guide
Learn how to train custom LoRAs for consistent character and style generation — no GPU required
Read GuideUpscale an Image Free
Enlarge images 2x/4x in the browser without losing quality
Read GuideAI Style Transfer: Ghibli & Watercolor
Turn a photo into Ghibli-style art with AI style transfer
Read GuideOld Video Enhancement Guide
Enhance old, low-quality video with AI — no download required
Read GuidePut These Prompts to Work
Open the LoRA image generator and try a prompt from this guide — from 20 credits per image. Imagera Typeset 3 and the edit models are one pick away in the same Sandbox.
