Ideogram 4.6 on Segmind: generate and precisely edit images with exact text

Ideogram 4.6 generates posters, logos and photoreal images with exact in-image text at native 2K or 4K. Its Precise Edit endpoint adds legible text or swaps one region while the rest stays unchanged.

Ideogram 4.6 is Ideogram's frontier image model. Its signature is the one thing most generators still get wrong: exact, legible text inside the picture. It renders headlines, logos and posters with quoted words spelled correctly, and generates or edits at native 2K or 4K with no upscaling. Realism, texture and color accuracy also improve over the previous generation. On Segmind it ships as two endpoints: text to image turns a prompt into a finished image, and precise edit changes one region while the rest stays unchanged.

Text to image

The Generate endpoint takes a text prompt and nothing else. Lead with the subject and quote the exact words you want rendered near the start, then pick a size, a quality tier and a background. It is built for design work where the copy has to be right: posters, logos, album covers, packaging and multilingual ad creative. Both quoted lines in the poster below come back spelled correctly and cleanly set.

ModelIdeogram 4.6Render exact in-image text and logos at native 4K.
In-image typographyfor Ideogram 4.6
A vintage letterpress travel poster for a mountain railway. Bold headline text reads 'IDEOGRAM 4.6' and a smaller line reads 'NATIVE 4K'. Warm cream paper texture, deep teal and burnt orange inks, a steam train crossing a stone viaduct at sunrise.
Try it
  • Example output

Both quoted lines come out spelled correctly and cleanly set inside the layout, so quote each line and keep it short.

Native 4K detailfor Ideogram 4.6
An extreme macro photograph of a peacock jumping spider perched on a dew-beaded fern frond. Its iridescent sapphire-and-amber scales, fine leg hairs and two large forward-facing eyes are in razor focus, and each dew droplet catches a bright reflection, under diffused overcast daylight against a blurred green background.
Try it
  • Example output

At size auto_4k the frame holds hair-fine texture edge to edge with no upscaling, so expect a larger file.

Transparent background (new in 4.6)for Ideogram 4.6
A cut-out studio product photo of a cherry-red vintage Italian motor scooter in full side profile, with chrome mirrors, a round headlamp and a tan leather seat, isolated on a fully transparent background with clean cut edges and no backdrop.
Try it
  • Example output

Setting background to transparent returns an alpha PNG you can drop straight onto any layout, ideal for logos and stickers.

Multilingual textfor Ideogram 4.6
A photoreal dusk photograph of a whitewashed flamenco venue facade in Seville. A glowing red neon sign near the top reads 'TABLAO EL DUENDE', with a smaller amber neon line below reading 'FLAMENCO EN VIVO'. A wrought-iron lantern hangs beside the arched wooden doorway.
Try it
  • Example output

Both Spanish lines render verbatim, so set magic_prompt to off when wording in another language must stay exactly as typed.

Use text to image from the API

Send a prompt and the settings you want, and the response returns the finished image.

APIGenerate a poster from a promptAPI reference
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind

result = segmind.run(
    "ideogram-4.6",
    prompt="A vintage letterpress travel poster for a mountain railway. Bold headline text reads 'IDEOGRAM 4.6' and a smaller line reads 'NATIVE 4K'. Warm cream paper texture, deep teal and burnt orange inks, a steam train crossing a stone viaduct at sunrise.",
    size="auto",
    quality="medium",
    background="auto",
    magic_prompt="auto",
    enable_copyright_detection=False,
)
print(result.get("output"))
# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/ideogram-4.6" \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "prompt": "A vintage letterpress travel poster for a mountain railway. Bold headline text reads '\''IDEOGRAM 4.6'\'' and a smaller line reads '\''NATIVE 4K'\''. Warm cream paper texture, deep teal and burnt orange inks, a steam train crossing a stone viaduct at sunrise.",
  "size": "auto",
  "quality": "medium",
  "background": "auto",
  "magic_prompt": "auto",
  "enable_copyright_detection": false
}' | jq -r .request_id)

# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
  -H "x-api-key: YOUR_API_KEY"

# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
  -H "x-api-key: YOUR_API_KEY"

Precise edit

The Precise Edit endpoint takes an image and a prompt that names only the change. It copies every pixel outside the edit straight from the source, so you can add a headline, swap an object or restyle one area without re-rendering the whole picture. Add a mask or a context_window for the tightest region control, and up to four reference images to bring in a product, logo or pattern. The result comes back at the input's exact width and height.

ModelIdeogram 4.6 EditEdit image regions precisely with legible text, rest unchanged
Headline set into the bannerPoster with a blank bannerInputOutput
A headline written into the poster's blank banner, the illustration untouched

The headline 'IDEOGRAM 4.6' and sub-line 'NATIVE 4K' land cleanly in the blank banner while the train, viaduct and art-deco border stay identical.

  • Plain canvas tote
    Input
  • Block-print swatch
    Reference
  • Patterned tote
    Output
A reference pattern printed across a plain tote, handles and wall unchanged

Pass the motif as a reference image and name only where it should land, up to four references or three alongside a mask.

Cobalt-reglazed bowlSpeckled stoneware bowlInputOutput
A 3456x2592 bowl reglazed and returned at the same size

A 3456x2592 input edits at 4K and returns at the same size, with the clay foot ring and oak grain untouched.

Use precise edit from the API

Send the source image, a prompt describing only the change, and any mask or reference images.

APIWrite a headline into a poster bannerAPI reference
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind

result = segmind.run(
    "ideogram-4.6-edit",
    image="https://segmind-resources.s3.amazonaws.com/input/ideogram-4.6-edit-poster-src.jpg",
    prompt="Set the headline 'IDEOGRAM 4.6' in bold condensed vintage letterpress capitals across the blank banner at the top, with a smaller sub-line 'NATIVE 4K' beneath it in spaced caps, inked in deep teal; keep the steam train, the stone viaduct, the snowy alpine peaks, the cream paper texture and the art-deco border unchanged",
    mask="",
    reference_images=[],
    quality="medium",
    context_window="",
    seed="",
    enable_copyright_detection=False,
)
print(result.get("output"))
# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/ideogram-4.6-edit" \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "image": "https://segmind-resources.s3.amazonaws.com/input/ideogram-4.6-edit-poster-src.jpg",
  "prompt": "Set the headline '\''IDEOGRAM 4.6'\'' in bold condensed vintage letterpress capitals across the blank banner at the top, with a smaller sub-line '\''NATIVE 4K'\'' beneath it in spaced caps, inked in deep teal; keep the steam train, the stone viaduct, the snowy alpine peaks, the cream paper texture and the art-deco border unchanged",
  "mask": "",
  "reference_images": [],
  "quality": "medium",
  "context_window": "",
  "seed": "",
  "enable_copyright_detection": false
}' | jq -r .request_id)

# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
  -H "x-api-key: YOUR_API_KEY"

# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
  -H "x-api-key: YOUR_API_KEY"

Side by side

The same launch poster is built two ways here.

  • Generated from a prompt
    Text to image
  • Headline edited into a blank banner
    Precise edit
The IDEOGRAM 4.6 / NATIVE 4K poster, generated from a prompt and edited into a blank banner

Text to image draws the whole poster, type included, from one prompt; precise edit writes the headline into an existing blank banner and leaves the illustration pixel-identical.

Shared settings

SettingValueMembersWhen to use
prompttext; quote exact words near the startbothevery run; on precise edit, name only the change
qualitymedium (default), high, very_lowbothhigh for dense or small text; very_low for quick 2K drafts
sizeauto, auto_2k, auto_4k, or WxH on a 32-px gridtext to imageauto_4k or any side over 2048 px for native 4K
backgroundauto (default), transparenttext to imagetransparent for an alpha PNG (logos, stickers)
magic_promptauto (default), on, offtext to imageoff to keep wording exactly as typed
imageJPEG, PNG or WebP sourceprecise editrequired; its size sets the output (over 2048x2048 edits at 4K)
maskblack-and-white image, input sizeprecise editblack marks the area to edit; gives the lowest drift
context_windownone, auto, or a pixel boxprecise editto edit a sub-region of a large source at full resolution
reference_imagesup to 4 (3 with a mask)precise editto bring in a product, logo, pattern or style
seed0 to 2147483647bothfix a value to reproduce a result

Good to know

  • Very dense or very small body text can still show occasional errors, so use quality high for dense, small type.
  • Non-Latin scripts work but are still expanding; you get the cleanest results in the target language with short quoted text.
  • Text to image is prompt-only and expands a short prompt by default, so set magic_prompt to off for verbatim wording; image and reference inputs go to precise edit.
  • On precise edit, whole-image edits can drift more than masked ones, so add a mask or context_window for tight control, and the output always matches the input's exact size.

Resources

  • #ideogram-4-6
  • #ideogram
  • #model-launch
  • #text-to-image
  • #text-rendering
AM

Abhilash Mirji

@abhilashmirji

Product engineer at Segmind, shipping new AI models faster than my cats can knock things off the desk. 11+ years turning ambiguous problems into dependable products. Off-call: drumming, scuba diving, and getting outvoted by four cats and two dogs.

View profile

Posts

More from Abhilash Mirji

Make your own with Segmind

Run every model and Pixelflow in this post, and build your own workflows.

Start creating