Ideogram on X

The vector model carries the section: 67 of its 107 people describe turning a prompt, or an uploaded image through Vectorizer, into an SVG whose shapes and letters stay editable. Files open in Illustrator or Figma without retracing, logos and posters scale without blur, and people name logos, book covers, posters and technical diagrams as what they made.

Plain enthusiasm is next, 22 people praising Ideogram 4.5, the new app UI and video integration, calling it a banger release and in one case making it a default editing option for their users.

13 people praise a different thing: edits that stay clean over many rounds, ten, twenty, thirty in a row, without drift or artifacts building up; one says models usually start drifting around edit 15. 6 rank it best, among them the most precise editing model and number one on an edit-fidelity benchmark. 6 are mixed or critical: output "not at all accurate", Arabic letters correct but diacritics unstable, too many rejected images, and doubt about the "first vector model" claim.

The misses people name are concrete: a generation that is "not at all accurate", Arabic text whose letters are right but whose diacritics are unstable (tested on one sentence and four images, the poster says it is not a final verdict), too many rejected images, and a hope that the new model will not refuse to draw things the way v4.0 did. One person trying the vector output could not get an SVG without a background: the first try came back with a black rectangle, the second with a fake "transparent" texture as vector.

The other complaint is about timing, from one person: that the 4.5 open weights are late enough that Flux 4 and Qwen Image 3 will land first and take the attention Ideogram is waiting for.

Price does not come up in this section.

Three of the four people weighing Flux against Ideogram come out for Ideogram: 4.5 called "far superior to Flux3", more accurate than Flux 3 on an image-editing drift benchmark, and a tutorial on editing without quality loss that also covers Flux 3. The fourth prefers Flux 3 for edits because it preserves the photo better, "especially faces".

QuiverAI is the vector comparison: one person asks whether Ideogram has just ended QuiverAI's business model and wants access to try its image-to-vector output, another asks whether it beats QuiverAI, which "did a poor job of vector conversion for me".

Single people name GPT Image as another model they prompt with for layout and detail control, using Ideogram's Magic Prompt to specify layouts; Gemini as what they turned to for image-to-vector conversions, "pretty decent"; and Qwen, expected to ship before Ideogram's open weights and take the attention.

The early-access posts are short and often hand over a username: people ask to be let in to try the vector model, some saying which Ideogram models they already use in their own products. One asks in the same breath about image-to-vector output and whether the SVG is well optimised.

Openness is the next question, 5 people: whether the model will be open or closed source, when the 4.5 open weights arrive, whether a version runs locally.

4 ask about precise control of what comes out: how exact text and responsive layouts are, whether there are multi-turn edit examples on images with text, and whether a single shape's colour can be changed without regenerating the whole image.

3 ask how the vector model copes with harder art, fragmented brush strokes and intricate patterns, and whether files open in Illustrator or Figma as clean separated shapes rather than "thousands of messy nodes". Two people ask about reach: an API with EU GDPR compliance and no training on their images, and when the model arrives in Magnific.

Two people are waiting on the open weights, one saying they cannot wait, the other waiting on "the open model for SS".

The other three wishes are one person each. Someone wants the free app on Huawei phones, not just Android. Someone else, hoping for DXF, still calls real vector output "amazing progress". And someone who would use the SVG output for a set of small icons wants to change a single colour directly in the file rather than generating the whole image again.

The wish for open weights is the same ask that shows up in questions, so it is worth reading alongside those: the release date, and whether the model can run locally, are the two things this product's keenest users keep asking for.