New voices, GPT Image 2.6 GIFs and updated Claude and Gemini
Seven new speech models, GPT Image 2.6 GIF and SVG output, 15-second Omni 1.2 clips, Claude Fable 5.2, GPT-6.1 Terra and Gemini 4 Carbon.
Seven new speech models are now available in NetRoom. This update also brings new OpenAI, Google and Anthropic text models, GPT Image 2.6 image generation and Gemini Omni Flash 1.2 video. They share one account and one balance.
Seven new text-to-speech models
- Google Gemini 3.8 Flash TTS and Flash Lite TTS: 30 preset voices each, two-speaker dialogue and a separate field for delivery instructions.
- Microsoft MAI-Voice 2.1 Flash and MAI-Voice 2.1: preset voices across locales, including Russian Grant, Harper, Lev and Masha. The selected voice determines the language.
- Mistral Voxtral TTS 3: preset voices with emotion and pacing tags in the text.
- AssemblyAI TTS 1: another option for narration and voice applications.
- Inworld Realtime TTS-2: delivery tags, preset voices and speed control, including four Russian voices.
Short demos let you audition voices before generating. Controls vary by model. For Gemini TTS, use the separate “Voice direction” field for instructions such as “warm delivery, calm pace, a short pause before the last sentence.” Keep the spoken words in the main text field.
Generate speech with your own voice
Voice cloning is available with Fish Audio S2.1 Pro, Fish Drama 3, HeyGen Voice and HeyGen Voice Turbo. In Audio or Studio, choose “Clone your voice”, upload a recording or record a sample, then enter the new text to generate.
A ready-to-read recording script fills the transcript automatically. HeyGen also accepts samples without a transcript. Listen to or replace your sample, then reuse it: the sample is remembered in this browser for your account and stays when you switch compatible models.
Use your own voice or a voice you have permission to use. Review pronunciation and similarity after generation; results depend on the recording and model.
GPT Image 2.6 adds GIF and SVG
GPT Image 2.6 Flare and Sunburst replace their 2.5 versions. Alongside JPG, PNG and WEBP, you can now select GIF or SVG.
To make a GIF, describe one short looping action and choose GIF as the output format. The current mode generates 9 frames at 0.2 seconds each, looping indefinitely. Frame dimensions are fitted automatically up to 960 pixels per side. GIF and SVG generation use text only, without reference images.
This GIF was generated while verifying the update:

Omni Flash 1.2: clips up to 15 seconds
Gemini Omni Flash 1.2 generates video with native audio, up to 15 seconds from text or one first frame. Reference inputs and first/last frame pairs support up to 10 seconds. Output resolutions reach 4K.
New text models
Claude Fable 5.2 replaces Fable 5. The catalog also adds GPT-6.1 Terra, Terra Pro and Gemini 4 Carbon.
Terra is an option for everyday coding, writing and agentic tasks. Terra Pro uses stronger reasoning: it has the same token rates but can take longer and consume more tokens. Fable 5.2 and Carbon expand the choices for demanding analysis, coding and document work.
Pricing
Speech catalog prices now reflect current provider rates. Most speech models charge by input text volume; Gemini TTS charges for input text tokens and output audio tokens. The price shown before generation is an estimate, with the final amount based on the generated result.
The new models are available now. Browse the model catalog or start generating.