First-run guide
This guide shows how to use chutes without assuming a particular model or interface layout. Start with one small task, inspect the result, and adjust your prompt before trying anything more complex.
You can begin with a short text task. Check the available controls before assuming a model or setting will be present.
Keep your first attempt small enough that you can tell whether the response answered the request.
Follow the action above to open the tool. Find the prompt field and note any visible model selector or task settings. Use an available option suited to text if your first task is a written question; do not assume every model handles the same inputs.
State the task, supply only the necessary context, and specify the output you want. For example, ask for three bullet points summarizing a passage you provide, rather than asking the model to summarize an unspecified document.
Compare the response with your prompt. If it missed a constraint, identify that constraint in a follow-up or submit a revised prompt. Verify factual claims, calculations, and code independently before using the output.
The most useful habit is to change one part of the request at a time, then compare what changed in the answer.
1
In Chutes, a prompt such as “Help with my writing” leaves the audience, length, and purpose open to interpretation. A more useful request names the material and the result: “Rewrite this paragraph for a nontechnical reader in under 100 words, preserving the figures.” Paste the paragraph with the request. If you do not have source material yet, ask for an outline rather than a summary of something the model cannot see.
2
Read the model name and any input guidance shown in the interface. A model offered for text may not accept an image, and a different model may produce a different style of answer to the same request. For your first Chutes test, keep the prompt unchanged while trying another available model. That gives you a clearer comparison than changing the model, prompt, and settings together.
3
A confident response is not proof that its facts are right. Check quotations against the source, run suggested code in a safe environment, and confirm any consequential claim with an independent reference. If the answer is too broad, ask Chutes to identify assumptions or to return a shorter, structured version. Do not paste private information merely to make the request more detailed.
A current browser with a working connection — If the page fails to load, test another site before troubleshooting the prompt.
One small, specific task you can evaluate — A short rewrite, explanation, or outline is easier to judge than an open-ended project.
The text needed to answer your questionoptional — Provide source text when requesting a summary or rewrite; omit unnecessary personal details.
A way to verify the result — Keep the original source, expected format, or a test case nearby.
A second available model to compareoptional — Useful for comparison, but not necessary for a first prompt.
These related pages cover model selection, chat use, and a different tutorial path if that is where you need help.
A failed attempt does not always mean the prompt is wrong. Isolate the problem before repeating the same request.
Chutes can return fluent text that omits a requirement or states something inaccurate. This guide cannot make model output authoritative.
What to do instead
Ask for a concise answer with explicit assumptions, then check important claims against a reliable source.
A pasted file path, a private webpage, or an image description is not the same as supplying content the selected model can access.
What to do instead
Check the visible input options. Paste relevant text directly when appropriate, or choose an available model that explicitly supports the input.
A blank page, stalled response, or repeated error may be caused by the connection, browser, or service rather than your wording.
What to do instead
Refresh once, test a short plain-text request, and try again later if the issue persists.
Mixing several goals, formats, and source passages into one first prompt makes it difficult to see which instruction was missed.
What to do instead
Split the work into smaller prompts and check each result before continuing.
Once a short Chutes request works, add one constraint at a time: the intended reader, a length limit, or a required output format. Save the original prompt and compare each result with it. For research, ask the model to separate supplied facts from assumptions; for code, request a small test case. These habits help you spot improvements without mistaking a more polished answer for a more accurate one.
Try a short task with an answer you can evaluate, such as rewriting a paragraph you supply for a specific audience. State the desired format and compare the result with your source before trying a larger task.
Use the model controls that are visible in the interface, if any, and check whether the selected option supports your input. If a model is already selected, you can begin with a small text prompt and compare other available options afterward.
The request may be missing source material or a clear description of the output. Rewrite it as a single task with the necessary context and one or two concrete constraints, then compare the new response with the first.
Check your connection, refresh once, and try a short plain-text prompt. If that also fails, the issue may not be your prompt; avoid repeatedly submitting sensitive material while troubleshooting.
No. Check facts against independent sources, inspect quoted material against the original, and test code before using it. Treat the response as a draft, especially when an error could affect someone else.