For v2rayNG users seeing REALITY, SNI, and Flow: REALITY handles the secure handshake between your client and the proxy server, while Vision determines how eligible traffic is forwarded. First distinguish the node address, handshake name, and destination website; then check the relevant settings to narrow down whether the issue is connectivity, handshake authentication, or routing.
The REALITY handshake: node address vs. SNI
When you tap Connect, the Xray core in v2rayNG first connects to the node address and port in your configuration, then sends the configured SNI in the TLS handshake. The node address specifies where the connection goes; SNI specifies the server name presented during the handshake. They can be different. Putting the SNI in the address field changes the actual connection target—it is not just another way to write it.
The REALITY client derives a shared secret from ephemeral key material and the server public key in the configuration, then includes authentication data in the handshake. The server processes and verifies this data with the corresponding private key; it includes the Short ID. The client also verifies that the peer has the corresponding key material. SNI and the fingerprint affect how the handshake looks, but they do not replace key-based authentication.
- REALITY authentication succeeds: The proxy server completes the secure connection handshake with the client. Only then does the VLESS session begin, where the user's identity is checked and proxy requests are handled.
- REALITY authentication fails: With a valid configuration and a reachable destination, the server can forward the connection to the configured destination, which then handles a standard TLS handshake. This path does not grant access to the VLESS proxy.
So “using a real website” does not mean that the website encrypts traffic for authenticated clients or lends them its private key. The website used for the handshake and the website you ultimately visit are two different things. Opening the relevant hostname in a browser and seeing its certificate or page only confirms that browser request worked; it does not prove the proxy handshake succeeded.
XTLS Vision: which encryption layer is removed?
When you browse an HTTPS site, for example, your browser already has a TLS session with the website. Standard encapsulated proxying adds another secure outer connection between the client and proxy server, encrypting the inner TLS records again as ordinary bytes. Once the proxy server removes the outer layer, what remains is still ciphertext between the browser and website—not readable webpage content.
- At the start of the connection: Vision handles the initial data inside the secure outer connection and adds padding to some early traffic. Both ends identify the inner TLS handshake and record boundaries as traffic is forwarded.
- When direct forwarding conditions are met: If both ends support the relevant flow and detect eligible inner TLS 1.3 traffic, subsequent data can switch to direct forwarding of the inner TLS records, without encrypting and decrypting each record in the outer layer.
- When direct forwarding conditions are not met: A lock icon in the browser address bar does not mean every connection has switched to direct forwarding. Traffic that does not meet the requirements continues through the outer-layer-protected path; enabling Vision does not make ordinary plaintext HTTP traffic go directly in the clear.
What Vision removes is redundant processing along the proxy path, not end-to-end encryption between the browser and website. Direct forwarding may also reduce copying between encryption and forwarding buffers, but it does not mean zero-copy. Whether the system can use a more copy-efficient path depends on the core implementation, operating system, and connection type. The presence of xtls-rprx-vision in the configuration alone does not prove that zero-copy is enabled on Android.
Check your v2rayNG settings: security and Flow are separate
Open the edit screen for the relevant node from the v2rayNG main list. First confirm that you selected the VLESS node you just imported, then check its transport, security type, and Flow. The edit menu and field labels may vary by client version, so use what appears on your node's edit screen. The two cards below list separate parts of the same connection, not mutually exclusive configurations.
REALITY handshake
- Security type
reality- Server name
- Enter the SNI accepted by the server
- Public key
- Use the public value corresponding to the server's private key
- Short ID
- Use the hexadecimal value accepted by the server
- Fingerprint
- Enter the value specified by your configuration provider
Get the public key and Short ID from the same configuration. Never enter the server's private key on the client.
VLESS and Vision flow control
- Protocol
VLESS- User ID
- Must match the corresponding user in the server configuration
- Transport
- A common direct-connection setup uses TCP, sometimes labeled RAW in the interface
- Flow
xtls-rprx-vision; the corresponding server-side user must use matching flow control
Flow is not a transport type and cannot replace the node address, port, or user ID.
A successful subscription import only means the client was able to read the configuration; it does not mean the server is using the same settings. If a node stops connecting after a subscription update, compare the old and new address, port, and security settings first. Changing the SNI, public key, or Flow yourself alters multiple troubleshooting variables and makes it harder to identify the cause.
Ports and performance: verifiable figures, not speed promises
Use the server's actual listening port. The public key's raw byte length is not the same as its encoded text length; do not truncate it to 32 characters. TLS 1.3 also needs context: the version negotiated between your browser and the destination website is different from the outer handshake between the client and proxy server.
Reducing redundant encryption mainly lowers per-byte processing overhead. It does not shorten the physical network path or automatically eliminate packet loss. First-page load time also depends on DNS resolution, the proxy handshake, and the website's response. A single request's latency on a small page does not represent sustained download throughput or phone battery use.
Bottom line: compare performance on the same connection
If you manage the server, run a comparison using the same device, network, server, and large file. Use matching configurations on both ends for each setup, test each three times, and compare median throughput; record the first-request time separately. If the two setups use different exits or the network changes during testing, you cannot attribute the difference to Vision.
Connection failures and pages that won't load: diagnose before changing settings
Repeated timeouts—should you change the SNI first?
First check the dial address and port on the node edit screen. If the address is a hostname, make sure it resolves on your current network. Check the first failure in the logs: if it reports a TCP connection timeout or refusal, investigate the server listener and network reachability first, and leave the SNI unchanged for now.
The port is reachable, so why is the handshake still failing?
A response from the port only confirms that a TCP connection can be established. Check the SNI, public key, Short ID, and fingerprint against the values from your configuration provider. On Android, search Settings for “Date & time” and enable automatic time; if the server enforces a handshake time-difference limit, check its clock too.
It says connected, so why can't one app open websites?
A VPN connected status only means the local VPN has started. First use a browser known to be routed through the proxy to open a known-accessible HTTPS page. Then check Settings → Per-app proxy to confirm whether the affected app is included or bypassed. If only certain domains fail, check their matching routing rules and DNS results.
It works after clearing Flow—is Vision broken?
First check the Flow configured for that VLESS user on the server and whether the client core supports the corresponding combination. Clearing the field changes the flow control used for requests. A change in connectivity only shows that the two configurations behave differently; it does not by itself prove Vision is faulty. When restoring the setting, make sure both ends use matching configurations.