Skip to content

smixs/visual-skills / image

Image prompting skill for Nano Banana (NBP/NB2) and GPT Image 2.5 (Flare/Sunburst). Writes ready-to-use prompts with model/quality/size recommendations. Use when: "нарисуй", "сгенерируй картинку", "image prompt", "промпт для картинки", blog covers, slides, posters, product shots, UI mockups, storyboards, character sheets, edit/colorize, style transfer, vision analysis, image-to-prompt, nb, NBP, NB2, gpt-image-2.5, multi-panel grids, ecommerce product photography, fashion editorial, food/beverage ads, cinematic portraits. Do NOT use for: video (use video skill), 3D models, audio, non-image tasks.

references/de-slop.md

View on GitHub
# De-Slop — killing the "AI look"

Load this whenever the brief is **a real photograph**: portrait, reportage, product-in-hand, UGC, casting shot, anything that must pass as an unretouched capture. Also load it when the user complains that a result "looks AI", "too glossy", "not like the reference".

Do NOT load it for illustration, 3D, poster graphics, stylized art — there the polish is the point.

*De-slop* is the name of this whole discipline: casting, capture, located imperfection, grading. The booster-word hygiene inside it is the same rule GPT Image 2.5 already carries as §Anti-Slop in [gpt-image.md](gpt-image.md) — that file owns the model syntax, this one owns the discipline.

The premise: both model families carry a strong prior toward an idealized image. Left alone they beautify — skin loses pores, faces gain symmetry, clothes lose creases, backgrounds gain saturation. Anything in the reference that differs from that ideal is quietly discarded unless you escalate it.

## 1. The default prior — where beautification starts

Check every axis before writing. If the brief or reference differs from the middle column, the deviation must be escalated (§2).

| Axis | What the model reaches for | What a real reference usually has |
|---|---|---|
| Eyes | large, round, wide-open, symmetric | narrow, hooded, monolid, one eye smaller |
| Face | V-line jaw, doll proportions, mirror symmetry | long or flat midface, visible asymmetry |
| Skin | poreless, glowing, even tone | pores, peach fuzz, redness zones, uneven tone |
| Body | idealized silhouette | flat, square, heavy, or simply ordinary |
| Clothing | tighter, newer, more skin | the exact coverage and wear of the reference |
| Hair | volumized, glossy, every strand placed | flatter, clumped, flyaways, dull ends |
| Stance | model pose, chin up | the awkward real stance from the reference |
| Casting | idol / influencer look | an ordinary person |
| Background | saturated, tidy, gradient | muted, cluttered, one named hex |
| Extras | adds earrings, props, signage, text | explicitly absent |

Counter-steer toward the reference, not toward plainness. If the subject genuinely is glamorous, describe that faithfully — only the capture rules in §4-§6 still apply.

## 2. Escalate every deviation in two places

A deviation stated once, in passing, loses to the prior. State it twice:

1. **In the descriptive slot**, geometrically and measurably — "narrow eyes with a low single lid crease, iris about 60% covered", not "unusual eyes".
2. **In `Constraints`** (GPT Image 2.5 fifth slot, see [gpt-image.md](gpt-image.md)) — one imperative line per axis, labeled: `EYES:`, `SKIN:`, `BODY:`, `HAIR:`, `MEDIUM:`.

This is the one place where naming the unwanted version is legitimate — the `Constraints` slot exists for "what must NOT change or appear". Everywhere else keep the positive framing from [golden-rules.md](golden-rules.md) §2.

Nano Banana has no constraints slot: fold the same lines into one short closing sentence of the prose prompt.

## 3. Boosters banned from positive slots

These words push the prior harder, they do not add fidelity: `4K`, `8K`, `ultra-detailed`, `hyper-detailed`, `ultra-realistic`, `photorealistic render`, `masterpiece`, `best quality`, `sharp focus`, `crisp`, `intricate detail`, `flawless`, `stunning`, `perfect`.

On GPT Image 2.5 they actively degrade the result ([gpt-image.md](gpt-image.md) §Anti-Slop). On Nano Banana they are mostly inert, but they crowd out the facts that would have worked.

## 4. A capture pipeline instead of adjectives

Do not ask for "realistic". Name the device and the path the file took — that is what carries the texture.

| Medium | What to state | Residue to ask for |
|---|---|---|
| Video frame grab | grabbed from handheld footage, not a photo | soft detail, mild compression blocking, slight motion smear |
| Film scan | 35mm colour negative, lab scan | visible grain, halation around highlights, soft blacks |
| Phone photo | phone camera, auto HDR off | smeared shadow noise, faint edge halo, slightly flat tone |
| Mirrorless / DSLR still | natural light, no flash, no retouch | fine sensor noise, natural falloff, unpolished skin |

State the medium as a fact of the image, never as a render style: "a real phone photo", not "photorealistic render in phone style".

## 5. Imperfections, and where they sit

An imperfection without a location gets ignored or sprayed everywhere. Name the place:

- skin — "pores across the nose and cheeks, redness at the nostril wings, one mole below the left jaw"
- hair — "two flyaways above the crown, ends clumped on the right shoulder"
- fabric — "creases at the elbow bend, collar sitting slightly askew on the left"
- environment — "scuffed baseboard behind the subject, uneven paint on the wall"
- optics — "one blown highlight on the forehead, slight motion blur on the near hand"

Calibrate to the reference, but never set the count to zero. See the micro-detail checklist in [prompt-framework.md](prompt-framework.md) §Cinematic Verbose for the same discipline applied to hero shots.

## 6. Sharpness and grading

- Ask for no added sharpening and low micro-contrast. Never write "sharp focus" or "crisp" (see §3).
- Grading: flat or natural, muted-to-moderate saturation, low contrast — unless the reference clearly shows otherwise.
- Ban the glow family in `Constraints`: HDR, bloom, high clarity, luminous or radiant skin, glossy hair highlights, perfectly even background.
- On GPT Image 2.5 stay at `quality: medium` for this kind of work. The table in [gpt-image.md](gpt-image.md) §Quality Settings sends portraits to `high` — that entry optimizes identity fidelity, which here fights the brief: `high` sharpens micro-texture until it reads as retouching. Step up to `high` only when matching a supplied face outranks the unretouched look, then re-run this section's checks on the result.

## 7. Self-check before sending

- Every axis in §1 that deviates appears both in a descriptive slot and in `Constraints`.
- No booster word from §3 survives in any positive slot.
- A capture pipeline from §4 is named, and the medium is stated as fact, not as style.
- At least three located imperfections from §5, each with a place.
- Sharpness and grading clauses present (§6).
- Anything the model likes to add but the brief does not contain — glasses, earrings, a second person, signage, captions — is listed as absent.
- Nothing in `Constraints` contradicts a positive slot (asking for grain while banning grain).

## 8. When the result still misses the reference

Treat every mismatch as a variable that was under-specified, not as a bad roll. Split the offending description into finer sub-facts, add or sharpen its `Constraints` line, and re-run the same prompt — one change per iteration, per [golden-rules.md](golden-rules.md) §3. Re-rolling an unchanged prompt returns the prior.

---

*Author: Serge Shima ([t.me/aimastersme](https://t.me/aimastersme) · [sergeshima.com](https://sergeshima.com) · [aimasters.me](https://aimasters.me)) · License: CC BY 4.0 — attribution required · Source: [smixs/visual-skills](https://github.com/smixs/visual-skills)*
canvas-designanthropics/skillsCreate beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
docxanthropics/skillsUse this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files) or Word templates (.dotx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', '.dotx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx or .dotx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
pdfanthropics/skillsUse this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
pptxanthropics/skillsUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates (.potx), layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx or .potx filename, regardless of what they plan to do with the content afterward. If a .pptx or .potx file needs to be opened, created, or touched, use this skill.Stars: 198kUpdated Claude CodeCodexVerifiedInstall