How to invoke a JSON or file output in the next node & how to output a request as an array/object

[rantzo7]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:)
• Please do not modify this template :) and fill in all the required fields.

1. Is this request related to a challenge you're experiencing? Tell me about your story.

I try to get a JSON output when using LLM node with Json schema enabled but it keeps giving a text response (string) even if the schema and the prompt is giving the desired output.

And overall, how do i get the JSON output in any node when it's available as most of the time it allow only to use text variables for the next node in the flow.
It shouldn't be that hard to get the JSON or file output. Using tools like Make or zapier, it's very easy to get an array or object.

2. Additional context or comments

Here is the output i get even if JSON Schema is enebled. I expect a object or array response as a JSON or file field.

{
"text": "{"products":[{"name":"Grigora","url":"https://appsumo.com/products/grigora/"},{"name":"Nexter","url":"https://appsumo.com/products/nexter/"},{"name":"Biyond","url":"https://appsumo.com/products/biyond/"},{"name":"Phygital+", "url":"https://appsumo.com/products/phygital/"},{"name":"Bestever","url":"https://appsumo.com/products/bestever/"},{"name":"SyncSignature - Plus Exclusive","url":"https://appsumo.com/products/syncsignature/"},{"name":"TeliportMe Virtual Tours","url":"https://appsumo.com/products/teliportme-virtual-tours/"},{"name":"Dialora","url":"https://appsumo.com/products/dialora/"},{"name":"ThriveDesk","url":"https://appsumo.com/products/thrivedesk/"},{"name":"Charla","url":"https://appsumo.com/products/charla/"},{"name":"Currents","url":"https://appsumo.com/products/currents/"},{"name":"Sherlock Al","url":"https://appsumo.com/products/sherlockai/"},{"name":"Dorin AI","url":"https://appsumo.com/products/dorin-ai/"}]}",
"usage": {
"prompt_tokens": 5174,
"prompt_unit_price": "1.25",
"prompt_price_unit": "0.000001",
"prompt_price": "0.0064675",
"completion_tokens": 2480,
"completion_unit_price": "10",
"completion_price_unit": "0.000001",
"completion_price": "0.0248",
"total_tokens": 7654,
"total_price": "0.0312675",
"currency": "USD",
"latency": 21.336650199024007
},
"finish_reason": "STOP",
"files": []
}

[dosubot[bot]]

Hi @rantzo7! I'm Dosu and I’m helping the dify team.

Dify's LLM node outputs the model response as a string in the text field, even when JSON schema is enabled. This is by design for compatibility—downstream nodes expect text, not objects or arrays. If the model supports structured output and JSON schema is enabled, Dify will also include a parsed object in the structured_output field, but most workflow nodes only use the text field by default, so you see stringified JSON instead of a native object or array [source].

To use structured data in subsequent nodes, insert a Template node after the LLM node. The Template node can convert the JSON string in text or the object in structured_output into a format (string, number, or array[file]) that downstream nodes accept [source]. This workaround is necessary because Dify currently restricts node outputs to these types for compatibility.

File outputs are only populated if the model returns multimodal content (like images); otherwise, the files field remains empty.

Compared to Make or Zapier, which natively pass objects/arrays between steps, Dify's workflow currently requires manual conversion. However, comprehensive support for node-level structured output (object/array) is planned for reintroduction in version 1.2 [source].

For now, use Template nodes to bridge the gap and extract structured data for further processing or integration. If you need true file outputs, ensure your workflow or model is designed to produce them.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[rantzo7]

@dosu thanks for response. What i want to achieve is this:
1- get a JSON output from a scrapper
2- send this to LLM node (gemini) with json schema enable (expecting a JSON response as an array or object.
Used json schema:
{
"type": "object",
"properties": {
"products": {
"type": "array",
"items": {
"type": "object",
"properties": {
"name": { "type": "string" },
"url": { "type": "string", "format": "uri" }
},
"required": ["name", "url"]
}
}
},
"required": ["products"]
}
3. Getting a array/object output
4. Use JSON parser to parse/extract each value in the array
5. Using Iteration node to use each value in the array one by one and set it in a scrapper as an URL variable

Could you help me how to achieve that with dify? Or what is the best way to achieve that?

[dosubot[bot]]

You can achieve this workflow in Dify with these steps:

1. After your scrapper node outputs JSON, use a Template node to convert the JSON output (array/object) into a string, since the LLM node only accepts text or files as input. Reference the scrapper's JSON output in the Template node and output it as a string (source).
2. Pass the stringified JSON from the Template node into your LLM node with JSON schema enabled. Make sure your schema root is an object with an array property (like your products array)—Dify does not support arrays at the root level (source).
3. After the LLM node, use a Parameter Extractor (JSON parser) node. This node will scan the LLM output, extract the JSON, validate it against your schema, and output a structured array/object for downstream nodes (source).
4. Connect the Parameter Extractor node’s array output (e.g., products) to an Iteration node. The Iteration node will process each item in the array, exposing the current item as a variable (item) for use in downstream nodes (source).
5. Inside the Iteration node, use the item variable to set the URL or other fields in your next scrapper or HTTP node. Make sure to reference the correct variable path (e.g., {{iteration_node_id.item.url}}) (source).

A few caveats:

• If you use Gemini (Vertex AI Gemini Flash 2.0), JSON schema is not supported in Dify and will cause errors—disable JSON schema for this model (source).
• If you’re on Dify v1.2.0, Iteration nodes with LLM structured output arrays may crash the editor; upgrade to v1.3.1+ to fix this (source).
• If your LLM output includes markdown code blocks (backticks), you may need to strip them manually until server-side fixes are released (source).

This setup will let you pass structured data from a scrapper, through an LLM, parse it, and iterate over each item for further processing.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[rantzo7]

@dosu I followed these steps and I got it working.
Now I wonder how to get the output from the iteration within which I run a scrapper via https request.
I got the right output under the http request from every iteration.
These are the only output variable I can get in the next node: https://i.imgur.com/pPVMjql.png
I can't access the output from the scrapper request: https://i.imgur.com/CdwoSAJ.png
https://i.imgur.com/iQNERyb.png