Tag
SenseNova 6.8 的 Code2Video 展示了一系列通过代码生成的动画示例，包括重复几何阵列、字母字形以及科学图表场景，突出其在重复物体、文本标签和空间关系渲染上的能力。
MOSS-Image Turbo LoRA (4-step) v2.0 has been released on Hugging Face, now Euler-native so it runs on the stock flow-matching sampler in any standard LoRA loader with no CFG or custom pipeline files, delivering fast high-fidelity bilingual text rendering.
Vik Paruchuri investigates why vision-language models sometimes output CJK characters like 罪 instead of English words like 'vain', exploring script confusions in VLMs and how the issue was fixed.
TextVerifier 提出了一种针对中文文本渲染缺陷的逐字符验证方法，利用表意文字描述序列（IDS）作为文本塔输入，并在 TextRenderBench-ZH 基准上达到 0.844 的 F1，显著优于 OCR 置信度、通用 VLM 和 TextPecker，还可作为 Flow-GRPO 后训练的逐字符奖励进一步降低渲染错误。
InfoVis 是一个统一的 9B 幻灯片生成与编辑模型，在超过 1300 万条预训练记录和新提出的 VisSlideBench 基准上取得开源最佳表现，在文本渲染、指令遵循、视觉定位、设计和编辑等 11 项指标上领先所有开源模型，总分 7.94 接近 Nano-Banana-2 的 7.99，模型、代码、数据和基准均将开源。
UltraText Bench is a bilingual (English/Chinese) benchmark with 432 prompts across 24 scene categories for evaluating dense visual text rendering in image generation models, scored by the Q-Judge VLM across text fidelity, clarity, spatial quality, and scene quality — revealing strong trade-offs between model variants such as Z-Image-Turbo versus Z-Image-Base.
A deep-dive on fixing dotted-circle ("tofu") rendering issues in Myanmar text, part of Google's Noto font effort to provide comprehensive Unicode coverage.
VTR-Bench introduces a systematic benchmark for evaluating visual text rendering in video generation models, covering 300 prompts across five scenario categories, an automated evaluation pipeline with human alignment, and a Keyframe-Guided Agentic Framework. Experiments on 11 state-of-the-art models reveal widespread difficulties in rendering scene text, with the best model achieving a word error rate of 0.250.
SenseNova-U1.5-8B-MoT is a native unified multimodal model for enhanced visual creation, featuring improvements in image generation quality, text rendering, and precise control.
Higgsfield launched Layers, an image editor that decomposes any image into editable layers for text, subjects, and background, enabling independent edits without regenerating the whole image, with native 4K support.
Grok announces Imagine Image 2.0, a next-generation image model with precision editing, crisp text rendering, improved factuality, and real-world usefulness.
SenseNova released a preview of its U1.5 Lite model, showing benchmark gains in image generation and editing, with native 4K output and improved Chinese/English text rendering, though acknowledged weaknesses remain.
A developer investigates why footnote backlink glyphs render as emoji in RSS readers, explores Unicode variation selectors (U+FE0E and U+FE0F) for controlling text vs emoji presentation, and discusses the broader challenges of variation selectors for web content.
A TrueType font that uses OpenType rules to render QR codes from bracketed text inline, without separate image generation.
The article explores various fundamental problems with text rendering in terminal emulators, including character definition ambiguity, Unicode handling issues, flawed 2D grid assumptions, and cursor desyncs, highlighting the difficulty of supporting complex scripts and fonts.
A leaked video from Google's unreleased Gemini Omni model shows impressive text rendering on a chalkboard, but the model still struggles with consistency in other prompts. The model appears to be an extension of Veo and is expected to be officially announced at Google I/O.
OpenAI released an upgraded image model that keeps character appearance perfectly consistent across frames and renders crisp, stable text.
OpenAI introduces native image generation capabilities in GPT-4o, featuring improved text rendering, precise object handling (10-20 objects), and context-aware generation through conversational refinement. The model excels at practical, communicative imagery with accurate symbol rendering and integration with uploaded images.
A head-to-head comparison of four leading image generators — NanoBanana 2.1, ChatGPT Image 2.5, Flux 3, and Seedream 5.0 Pro — tested on marketing assets, targeted editing, and infographics using the same set of prompts: NanoBanana and ChatGPT tie in marketing scenarios, while ChatGPT Image 2.5 leads in complex text rendering and infographics.
ChatGPT Image 2.5 excels in complex image generation tasks, introducing features like templates and sketch-to-image conversion, but it still struggles with modifying complex text.